CN117042681A - Indirect sensing mechanism for cardiac monitoring - Google Patents

Abstract

The present disclosure relates to systems and techniques for detecting patient health changes based on patient data. In some examples, a medical system includes: a mechanical sensor configured to sense a first physiological parameter signal of a patient; and processing circuitry configured to: determining one or more characteristics of the first physiological parameter signal; applying a machine learning model to the one or more features of the first physiological parameter signal; and determining an estimate of the second physiological parameter based on the application of the machine learning model.

Description

Indirect sensing mechanism for cardiac monitoring

Priority claim

The present application claims the benefit of U.S. provisional patent application No. 63/163,607, filed 3/19 of 2021, the entire contents of which are incorporated herein by reference.

Government funding

The project of the present application was led to the sponsorship of European Union's horizons 2020 according to Marie Sklodowska-Curie) dial protocol No. 764738.

Technical Field

The present disclosure relates generally to medical systems, and more particularly, to medical systems configured to monitor physiological activity to learn about changes in patient health.

Background

Some types of medical systems may monitor various physiological data of a patient or a group of patients to detect changes in health conditions, such as notification of therapy delivery to improve the health condition of the patient. Physiological data may include, for example, cardiac Electrograms (EGMs), activity or motion, heart sounds or vibrations, oxygen saturation, and blood pressure. As an example, the medical system may monitor cardiac electromechanical function based on such physiological data and control delivery of therapy to improve cardiac electromechanical function, such as Cardiac Resynchronization Therapy (CRT), implantable Cardioverter Defibrillator (ICD) therapy, or another therapy for cardiac arrhythmias.

Disclosure of Invention

The medical systems and techniques as described herein provide real-time care in devices implanted in a patient's body and enhance the care with a monitoring mechanism that is non-invasive with respect to the monitored body part. In one example medical system, a medical device (e.g., a pacemaker) is implanted in a patient's chest to monitor cardiac activity, and in some examples, deliver therapy to correct abnormal cardiac activity in the patient. An exemplary medical system may utilize hemodynamic measurements for cardiovascular function of interest (e.g., cardiac synchrony) to define abnormal cardiac activity. A medical device (e.g., CRT device or ICD) may monitor electrocardiographic or mechanical activity to detect the degree of synchrony and, if desired, improve synchrony by delivering therapy such as defibrillation (i.e., shock treatment) from the CRT device or ICD. As another example, the medical device may monitor electrocardiographic or mechanical activity via hemodynamic sensing and deliver anti-tachycardia pacing (ATP) based on one or more criteria for certain hemodynamic measurements. In other examples, ICDs may employ hemodynamic sensing to participate in shock processing for life threatening arrhythmias or cardiac arrest. For both the degree of synchrony and the hemodynamic measurements, the sensor enables the medical device to capture signal data of the physiological parameter.

However, some physiological parameters, such as pressure in the heart chamber or in an artery or vein (e.g., an aorta or pulmonary vein), particularly indicative of the above-mentioned cardiovascular function of interest, often require that the sensor be invasively positioned at a location within the left side of the heart. Some example medical devices are configured to estimate Left Ventricular (LV) pressure or other left side cardiac mechanical function measurements, while other example medical devices are configured to estimate Right Ventricular (RV) pressure, left Atrial (LA) pressure, right Atrial (RA) pressure. One or more example medical devices may be configured to estimate pressure in one or more of four pulmonary veins or one or more of four coronary arteries entering the Left Atrium (LA). To minimize care related to interactions with a patient's heart, a medical device implements a technique as described herein that employs a mechanism that senses left or right heart function (such as left ventricular pressure) that is non-invasive with respect to the left or right side of the heart.

In the present disclosure, the medical device is configured with hardware to implement various machine learning techniques for estimating physiological parameters (such as left ventricular pressure) and then using the estimates in some functions. In one example, the pacemaker is configured with a machine learning model (e.g., neural network, decision tree, and other artificial intelligence/machine learning algorithms) that estimates left ventricular pressure measurements and their derivatives based on endocardial or epicardial mechanoreceptive signals (e.g., accelerometers). Pacemakers are relatively small devices due to the limitations and sensitivity of the human heart, and thus benefit from medical systems/techniques with a lower or manageable resource footprint. Given the computational complexity (or lack thereof) of the medical systems/techniques described herein, pacemakers may implement the machine learning techniques described herein without undue attention to any resource burden.

Examples of the machine learning techniques described above identify a set of features from the mechanoreceptive signals and apply a model (e.g., as a decision tree-based model) to estimate left ventricular pressure data. In the present disclosure, the model is designed and then calibrated/trained to provide reasonable predictions of left ventricular pressure measurements to record current or future points in time. An exemplary medical device implementing the machine learning techniques described above does not rely on directly sensing left ventricular pressure and maintains a desired level of accuracy to prevent errors, for example, when detecting/correcting abnormal heart rhythms. The sensor data generated by the medical systems/techniques of the present disclosure, including pressure measurements and any parameter data corresponding to the left ventricular pressure of the patient, is accurate and available automatically in real-time and upon request (if so configured). In view of the above, the present disclosure describes a technical improvement or technical solution integrated into a practical application.

In one example, a medical system includes a mechanical sensor configured to sense a first physiological parameter signal of a patient; and processing circuitry configured to: determining one or more features of a first physiological parameter signal, applying a machine learning model to the one or more features of the first physiological parameter signal; and determining an estimated value of the second physiological parameter based on the application of the machine learning model.

In another example, a method includes: determining one or more characteristics of a first physiological parameter signal sensed by the mechanoreceptors: applying a machine learning model to the one or more features of the first physiological parameter signal; and determining an estimated value of the second physiological parameter based on the application of the machine learning model.

In another example, a medical system includes a communication circuit coupled to a medical device via a network; and processing circuitry configured to: performing a training process on a corpus of cardiac activity data of a patient population or a subset of patients; generating a machine learning model based on the training process, wherein the machine learning model is configured to determine an estimate of the second physiological parameter using the first physiological parameter signal; and deploying the machine learning model for use in a medical device configured to monitor cardiac activity of a patient.

This summary is intended to provide an overview of the subject matter described in this disclosure. This summary is not intended to provide an exclusive or exhaustive explanation of the systems, devices, and methods described in detail in the following figures and description. Further details of one or more examples of the disclosure are set forth in the accompanying drawings and the description below. Other features, objects, and advantages will be apparent from the description and drawings, and from the claims.

Drawings

Fig. 1 illustrates an example environment of an example medical system in conjunction with a patient according to one or more examples of this disclosure.

Fig. 2 is a conceptual diagram of an IMD in the example medical system of fig. 1 according to one or more examples of the present disclosure.

Fig. 3 is a block diagram illustrating an exemplary configuration of the IMD of fig. 2 according to one or more examples of the present disclosure.

Fig. 4 is a functional block diagram illustrating an exemplary configuration of the external device of fig. 1 according to one or more examples of the present disclosure.

Fig. 5 is a block diagram illustrating an example system including an access point, a network, an external computing device (such as a server), and one or more other computing devices that may be coupled to the medical device and the external device of fig. 1-4, according to one or more examples of the present disclosure.

Fig. 6A is a flow diagram illustrating a first stage of a machine learning technique that enables less invasive monitoring of physiological parameters in accordance with one or more examples of the present disclosure.

Fig. 6B and 6C are flowcharts illustrating, respectively, a second stage and a third stage of the machine learning technique of fig. 6A in accordance with one or more examples of the present disclosure.

Fig. 7A is a flowchart illustrating a first operation of a medical system running on a cloud computing environment according to one or more examples of the present disclosure.

Fig. 7B is a flow chart illustrating a second operation of the medical system of fig. 7A according to one or more examples of the present disclosure.

Like reference characters designate like elements throughout the description and figures.

Detailed Description

The present disclosure describes systems and techniques for estimating certain data corresponding to a patient's heart, including the left ventricle. The present disclosure, in which implantable medical devices are described as exemplary embodiments, presents a mechanism to support these devices to estimate patient cardiac activity that is non-invasive with respect to the portion of the heart (such as the left ventricle) whose activity is estimated. The present disclosure sets forth solutions or improvements to current medical devices that typically encounter difficulties when sensing portions of a patient's heart (e.g., the left or right ventricle, the left or right atrium, or other portions of the left or right side of the heart).

According to some examples, the present disclosure describes a (mathematical) model that is calibrated/trained to accurately estimate pressure data (e.g., left ventricular pressure or its derivatives) without directly sensing (e.g., measuring) pressure data from the patient's heart. Conventional techniques that rely on direct sensing employ an invasive mechanism to acquire sensor data via direct sensing.

The present disclosure relates to machine learning techniques (e.g., modeling techniques including training processes and evaluation processes) that provide (at least in part) accurate estimates of left ventricular pressure. Some example machine learning techniques generate a decision tree-based model for estimation, but these techniques are not limited to this model structure and other suitable models may be employed. In fact, most, if not all, of the model structures are suitable for estimation. When the model is fully configured and then sufficiently trained to estimate left ventricular pressure and one or more derivatives, some example machine learning techniques implement the model into a medical device.

The implantable medical devices described herein implement the above-described models (e.g., in detection logic) to improve performance of device functions and/or quality of medical care for a physician, particularly where the function of one or more medical devices is to be improved by left ventricular pressure data. A pacemaker is one example of the implantable medical devices described above, and the pacemaker may use accurate left ventricular pressure data to correct abnormal rhythms and/or improve cardiac synchrony. Instead of direct sensing, the pacemaker estimates left ventricular pressure measurements from the mechanoreceptive signal that are within a predetermined range (e.g., error). Because the model is fully trained and then programmed into the pacemaker, the estimated error of the pacemaker may be insignificant and represented by no performance impact or (e.g., statistically) insignificant performance impact.

The following description relates to a variety of medical devices (e.g., implantable devices, wearable devices, etc.) configured to monitor patient cardiac activity, detect mechanoreceptive signals and other input data at least some of which correspond to the left ventricle of the patient, and estimate contemporaneous pressure measurements in the chamber. These medical devices implement machine learning techniques as described herein, and in some examples, function with some capabilities for medical systems as described herein. To illustrate by an exemplary medical system, a cloud-based network device may utilize the marker data and learning algorithms to train a model and, when the model is sufficiently trained, to correctly or accurately correlate features (e.g., including mechanoreceptive signals captured over a period of time) with changes in left ventricular pressure.

In this way, the systems and techniques of the present disclosure may advantageously enable improved accuracy in detection of diseases that double-fold a patient and negatively affect the patient's health condition, and thus better treat and/or correct the patient's disease, resulting in an improved condition of the patient's condition.

Fig. 1 illustrates an exemplary medical device system 10 in conjunction with a patient 14. The medical device system 10 is an example of a medical device system configured to implement the example techniques described herein for estimating physiological parameter values and, in some examples, controlling delivery of CRT to the heart 12 of the patient 14 based on the estimated physiological parameter values. In some examples, medical device system 10 includes an Implantable Medical Device (IMD) 16 in communication with an external device 24. In the illustrated example, IMD 16 may be coupled to leads 18, 20, and 22.IMD 16 may be, for example, an implantable pacemaker that provides electrical signals to heart 12 and senses electrical activity of heart 12 via electrodes coupled to one or more of leads 18, 20, and 22. In some examples, IMD 16 may have cardioversion and/or defibrillation capabilities.

Leads 18, 20, 22 extend into heart 12 of patient 14 to sense electrical activity of heart 12 and deliver electrical stimulation to heart 12. In the example shown in fig. 1, right Ventricular (RV) lead 18 extends through one or more veins (not shown), the superior vena cava (not shown) and Right Atrium (RA) 26, and into RV 28. A Left Ventricular (LV) coronary sinus lead 20 extends through one or more veins, the vena cava, the right atrium 26, and into the coronary sinus 30 to a region adjacent to the free wall of the LV 32 of the heart 12. Right Atrial (RA) lead 22 extends through one or more veins and the vena cava and into RA 26 of heart 12.

IMD 16 may sense electrical signals attendant to depolarization and repolarization of heart 12 via electrodes (not shown in fig. 1) coupled to at least one of leads 18, 20, 22. In some examples, IMD 16 may also sense electrical signals accompanying depolarization and repolarization of heart 12 via an extravascular electrode (e.g., an electrode positioned outside of the vasculature of patient 14), such as an epicardial electrode, an external surface electrode, a subcutaneous electrode, or the like. The configuration of electrodes used by IMD 16 for sensing and pacing may be monopolar or bipolar.

In one placeIn some examples, IMD 16 is configured to provide CRT to heart 12. In some examples, IMD 16 is configured to at least one of deliver fusion pacing to heart 12 and biventricular pacing to heart 12 as part of CRT. In some examples of fusion pacing, IMD 16 may deliver pacing stimulus (e.g., pacing pulses) to LV 32 via electrodes of lead 20, where the pacing stimulus is timed such that evoked depolarization of LV 32 is achieved in fusion with intrinsic depolarization of RV 28, resulting in ventricular resynchronization. In this way, pacing pulses (LV) delivered to LV 32 _P ) Conduction-delayed LV 32 may be pre-stimulated and help fuse activation of LV 32 with activation of RV 28 from intrinsic conduction. Depolarizing fusion of LV 32 and RV 28 may result in synchronous activation and contraction of LV 32 and RV 28. In the examples described herein, the fused pacing configuration may be referred to as "left ventricular" pacing. However, it should be understood that in any of the described examples, the fusion pacing configuration may include right ventricular pacing.

In some examples, when IMD 16 is in a biventricular pacing configuration, IMD 16 may deliver pacing stimulation (e.g., pacing pulses) to RV 28 via the electrodes of lead 18 and pacing stimulation to LV 32 via the electrodes of lead 20 in a manner that synchronizes activation and contraction of RV 28 and LV 32.

As discussed in further detail below, IMD 16 may be configured to adjust one or more pacing parameters based on the cardiac status of heart 12. In some examples, IMD 16 may be configured to adjust pacing parameters by delivering electrical stimulation therapy to heart 12 according to a fusion pacing configuration or a biventricular pacing configuration.

In some examples, CRT provided by IMD 16 may be used to maintain a heart rhythm of patient 14 having conduction dysfunction, which may result when the natural electrical activation system of heart 12 is disrupted. The natural electrical activation system of the human heart 12 involves several sequential conduction pathways starting from the Sinoatrial (SA) node and continuing at the atrial level through the barkmann's bundle and internode bundles, followed by the Atrioventricular (AV) node, the common bundle of his (Common Bundle of His), the right and left bundle branches, and finally distributed to the distal myocardial end via a Purkinje fiber network.

In a normal electrical activation sequence, the cardiac cycle begins with the generation of a depolarization wave at the SA node in the RA 26 wall. Depolarization waves are transmitted at the atrial level through the atrial conduction pathways of the barker bundle and the internode bundle to the LA 33 membrane. When the atrial depolarization wave reaches the AV node, the atrial septum, and the furthest walls of the right and left atria 26, 33, respectively, the atria 26, 33 may contract due to electrical activation. The aggregated right and left atrial depolarization waves represent P-waves of the PQRST complex of cardiac electrical signals, such as cardiac Electrograms (EGMs) or Electrocardiographs (ECGs). Atrial depolarization waves passing between a pair of unipolar or bipolar pacing/sensing electrodes located on or adjacent to RA 26 and/or LA 33 may be detected as sensed P-waves when their amplitude exceeds a threshold. The sensed P-waves may also be referred to as atrial sensed events or RA sensed events (RA _S ). Similarly, the P-waves sensed in LA 33 may be referred to as atrial sensed events or LA sensed events (LA _S )。

During or after atrial systole, the AV node distributes depolarization waves downward under the bundle of his of the septum within the ventricle. The depolarization wave may travel to the apical area of the heart 12 and then up through the purkinje fiber network. The aggregated right and left ventricular depolarization waves and the subsequent T waves that accompany repolarization of the depolarized myocardium may manifest as QRST portions of the PQRST cardiac cycle complex. A sensed R-wave may be detected when the amplitude of a QRS ventricular depolarization wave passing between a bipolar or unipolar pacing/sensing electrode pair positioned on or adjacent to RV 28 and/or LV 32 exceeds a threshold. Depending on the ventricle, the sensed R-waves may also be referred to as ventricular sensed events, RV sensed events (RV _S ) Or LV sensing event (LV) _S ) In this ventricle, the electrodes of one or more of leads 18, 20, 22 are configured to sense under certain conditions.

Some patients, such as those suffering from congestive heart failure or cardiomyopathy, may suffer from left ventricular dysfunction, thereby being damaged within the LV 32 by the normal electrical activation sequence of the heart 12. In patients with left ventricular dysfunction, the normal electrical activation sequence through the patient's heart is disrupted. For example, the patient may experience an intra-atrial conduction defect, such as an intra-atrial block. Intra-atrial block is a condition in which atrial activation is delayed due to conduction delay between RA 26 and LA 33.

As another example, a patient with left ventricular dysfunction may experience an intra-ventricular conduction defect, such as Left Bundle Branch Block (LBBB) and/or Right Bundle Branch Block (RBBB). In LBBB and RBBB, the activation signal is not conducted in the normal manner along the right or left bundle branch, respectively. Thus, in patients with bundle branch block, activation of RV 28 or LV 32 is delayed relative to the other ventricles, resulting in asynchrony between right and left ventricular depolarizations. This asynchrony can lead to reduced mechanical properties of the heart, which can be reflected in measurements such as ejection fraction, stroke volume, LV pressure, and derivatives of LV pressure.

CRT delivered by IMD 16 may help alleviate heart failure conditions by restoring synchronized depolarization and contraction of one or more chambers of heart 12. In some cases, fusion pacing of heart 12 as described herein enhances the mechanical properties of the patient's heart by improving the synchronicity of the depolarization and contraction of RV 28 and LV 32. In some examples, measurements of mechanical properties of the heart may be used to assess the performance of the CRT, and in some cases, as feedback for modifying one or more parameters of the CRT, such as the a-V interval or selection of which electrodes are used to deliver the CRT (or, alternatively, ICD shock treatment). However, measurements of mechanical properties on the left side of the heart tend to be invasive and thus do not facilitate long-term monitoring of CRT efficacy. The techniques of this disclosure may allow system 10 (e.g., IMD 16) to estimate such measurements to determine the efficacy of the CRT and allow feedback control of CRT parameters.

In some examples, IMD 16 also provides defibrillation therapy and/or cardioversion therapy via electrodes located on at least one of leads 18, 20, 22. IMD 16 may detect arrhythmias of heart 12, such as fibrillation of ventricles 28 and 32, and deliver defibrillation therapy to heart 12 in the form of electrical shocks. In some examples, IMD 16 may be programmed to deliver the progress of therapy (e.g., a shock with an increased energy level) until defibrillation of heart 12 ceases. In examples where IMD 16 provides defibrillation therapy and/or cardioversion therapy, IMD 16 may detect fibrillation by employing any one or more of the defibrillation detection techniques known in the art.

The external device 24 may be a computing device having a display viewable by a user and includes an interface to receive input from the user. In some examples, external device 24 may be a notebook computer, a tablet computer, a workstation, one or more servers, a cellular telephone, a personal digital assistant, or another computing device that may run an application program that enables the computing device to interact with IMD 16. The user interface may include, for example, a keypad and a display, which may be, for example, a Cathode Ray Tube (CRT) display, a Liquid Crystal Display (LCD), or a Light Emitting Diode (LED) display. The keypad may take the form of an alphanumeric keypad or a reduced set of keys associated with a particular function. External device 24 may additionally or alternatively include a peripheral pointing device, such as a mouse, via which a user may interact with the user interface. In some implementations, the display of the external device 24 may include a touch screen display, and the user may interact with the external device 24 via the display.

A user, such as a physician, technician, or other clinician, may interact with external device 24 to communicate with IMD 16. For example, a user may interact with external device 24 to retrieve physiological or diagnostic information from IMD 16. The user may also interact with external device 24 to program IMD 16, for example, to select values for operating parameters of the IMD.

For example, a user may use external device 24 to retrieve information from IMD 16 regarding the heart rate of heart 12, its trend over time, or the onset of an arrhythmia. As another example, a user may use external device 24 to retrieve information from IMD 16 regarding other sensed physiological parameters of patient 14, such as sensed electrical activity, posture, respiration, thoracic impedance, or other data related to the techniques described herein, from IMD 16. As another example, a user may use external device 24 to retrieve information from IMD 16 regarding the performance or integrity of IMD 16 or other components of system 10, such as leads 18, 20 and 22 or the power source of IMD 16. In such examples, physiological parameters of patient 14 and data regarding IMD 16 may be stored in a memory of IMD 16 for retrieval by a user.

The user may program the progress of the therapy using external device 24, select electrodes for delivering defibrillation pulses, select waveforms for defibrillation pulses, or select or configure a defibrillation detection algorithm for IMD 16. The user may also use external device 24 to program aspects of other therapies provided by IMD 16, such as cardioversion, CRT, or pacing therapies. In some examples, a user may activate certain features of IMD 16 by entering a single command via external device 24, such as pressing a single key or combination of keys of a keypad or a single point selection action with a pointing device.

Monitoring service 6, IMD 16, external device 24, and optionally another computing device (not shown in fig. 1) may communicate via wireless communication using any technology known in the art. The external device 24 may be configured to communicate with the external device via near field communication technology (e.g., inductive coupling, NFC, or other communication technology that may operate at a range of less than 10cm to 20 cm) and far field communication technology (e.g., according to 802.11 orRadio Frequency (RF) telemetry of a specification set or other communication technology that may operate at a range greater than near field communication technology). Examples of viable communication technologies may include, for example, radio Frequency (RF) telemetry, which may be an RF link established via an antenna according to bluetooth, wiFi, or Medical Implantable Communication Services (MICS), although other technologies are also contemplated. In some examples, external device 24 may include a programming head that may be placed in close proximity to the patient's body near the implantation site of IMD 16 in order to improve the quality or safety of communication between IMD 16 and external device 24.

The processing circuitry of IMD 16 (which may be combined with the processing circuitry of external device 24) may employ various techniques to capture physiological parameter signals over a period of time and subsequently analyze (e.g., parameter values encoded in) the captured physiological parameter signals to obtain an indicator of a non-slightly changing patient health condition including the heart health of patient 14. The present disclosure describes IMD 16 as having access to various hardware/software devices (e.g., sensors as components coupled to IMD 16 and/or IMD 16) for generating the above-described signals, including electrodes, accelerometers, pressure sensors, force transducers, and other sensors. The plurality of physiological parameter signals transmit information indicative of cardiac health and/or sensed cardiac activity of the patient 14, including electrical activity, pressure, movement/motion, and other environmental characteristics of body regions within and/or surrounding the heart 12. Examples of the above-described physiological parameter signals may include cardiac EGM or ECG (i.e., cardiac EGM signal signals), cardiac motion data (e.g., accelerometer signals), pressure data (e.g., pressure sensor signals), and the like.

IMD 16 may be configured with logic to implement techniques to estimate values of one or more physiological parameters of patient 14 based on characteristics of another physiological parameter acquired with less invasiveness. As described in this disclosure, the logic may employ a number of compatible mechanisms to successfully implement the techniques described above, such as mathematical or machine learning models (e.g., neural networks and/or decision trees), where each mechanism specifies criteria that may be used to estimate physiological parameter values.

In some examples, processing circuitry of medical system 10 analyzes patient data including data encoded in one or more physiological parameter signals (e.g., motion data of heart 12) representative of patient heart activity sensed by IMD 16 and may identify markers of heart disease or abnormalities (e.g., asynchrony). As described herein, directly sensing physiological parameters presents a number of difficulties for similar medical devices of IMD 16, such as having to configure sensors (e.g., pressure sensors) at (e.g., epicardially or endocardially) body locations (e.g., within a chamber of heart 12, such as the left ventricle). For purposes of illustration by way of an exemplary sensor for directly sensing pressure data of the heart 12 of the patient 14, fig. 1 depicts the pressure sensor 2 as being configured in an invasive position that may prevent some cardiac function.

As described herein, the processing circuitry of the medical system 10 enables indirect and thus less invasive sensing and monitoring of certain parameters of the heart health of the patient 14. For example, the processing circuitry of medical system 10 generates mechanisms that support estimating left ventricular pressure data without directly sensing such data. IMD 16 (as an example medical device, such as a pacemaker described herein) employs example machine learning techniques that in some cases use a trained model (e.g., a trained decision tree-based model) to calculate an estimate of left ventricular pressure data. The processing circuitry of IMD 16 may use these estimates to perform some function of the benefit to patient 14 (e.g., regarding cardiac health), and, if applicable, replace direct pressure sensor measurements. In some cases, IMD 16 does not include any hardware/software components for directly sensing the pressure within the left ventricle of the heart of patient 14, but rather relies on a training model for estimating sensed pressure data.

The present disclosure contemplates medical devices equipped with multiple hardware/software components to implement different exemplary techniques. An exemplary medical device (e.g., a therapy device including an implant such as IMD 16) performs a function that is beneficial and effective to the patient 14, which is facilitated by a number of configurable settings and direct sensing techniques to obtain physiological parameter data from a body region. The body of the patient 14 includes regions (e.g., organs such as the heart) that may be sensitive to any interactions, and directly sensing (current) conditions (e.g., left ventricular pressure) in those regions may increase the likelihood of adverse consequences for the patient 14.

Exemplary techniques may be used with IMD 16, which may communicate wirelessly with at least one of external device 24 and other devices not depicted in fig. 1. In some examples, IMD 16 is implanted outside of the chest of patient 14 (e.g., subcutaneously in the chest position shown in fig. 1). IMD 16 includes a plurality of electrodes (not shown in fig. 1) and is configured to sense a cardiac EGM via the plurality of electrodes and a plurality of other sensors (not shown in fig. 1) for sensing other physiological parameters, such as one or more accelerometer movements or vibrations. Although described primarily in the context of an example of a pacemaker in which IMD 16 is configured to deliver CRTIt is contemplated that the techniques described herein may be implemented in any implantable or external device to estimate one physiological parameter based on another physiological parameter. In some examples, IMD 16 employs a LINQ available from Medtronic, inc. Of Minneapolis, MN ^TM ICM, implantable or external defibrillator, intracardiac pacemaker, neurostimulator or drug pump.

The processing circuitry of medical system 10 facilitates training the decision tree-based model described above in environments having more resource capacity and capabilities than medical devices such as IMD 16. In one example, the processing circuitry of the computing system of the remote monitoring service 6 and/or the processing circuitry of the external device 24 performs another exemplary machine learning technique (which may be referred to as a training process) to generate a distribution of pressure measurements from the characteristic data. Typically, as some examples of the training model described above, the distribution represents the model described above and identifies a mapping between the characteristic data samples and the estimate for left ventricular pressure, for example, in a multivariate mathematical function or decision tree. Prior to the training process, the processing circuitry of medical system 10 may generate an untrained or pre-trained model and then cause the untrained or pre-trained model to be configured without a mapping (i.e., empty set) or with only a default mapping. Using the feature data samples and the reference pressure measurements as training data for the training process, the processing circuitry of medical system 10 generates an initial set of mappings, wherein the feature data samples are processed into an initial distribution (e.g., a linear distribution) of fitted reference pressure measurements. IMD 16 may be configured with a preclinical derived estimation algorithm to provide at least an initial profile and possibly a subsequent profile for further training/calibration. According to some examples, the training process may introduce a calibration step in which, over multiple iterations, the processing circuitry of medical system 10 updates the previous model with a (more accurate) distribution that better fits the reference pressure measurements. In some examples, the training process incorporates patient-specific calibration into the calibration step, wherein patient-specific reference data (e.g., (non) invasive pressure measurements, imaging data including echocardiographic data and electrocardiographic data, and/or the like) may be used to update the previous model and generate the training model.

Medical system 10 includes a computing system communicatively coupled to one or more medical devices including IMD 16 through a network connection and exchanging various data with IMD 16 via a wireless communication protocol. As specified by medical system 10, the computing system may be configured to operate a remote monitoring service 6 for IMD 16. The processing circuitry of medical system 10 may include one or more processors in the computing system of remote monitoring service 6 and may receive sensor signals storing sensed patient physiological parameters from different devices of IMD 16 via communication circuitry. At least some of these sensor signals are the result of (directly) sensing those parameters from (invasive or invasive) body locations (e.g., epicardial locations). The sensed physiological parameter (e.g., signal) includes actual sensor measurements that can be used as reference data for training a machine learning model. In one example, the remote monitoring service 6 may train an estimation model using the received patient physiological parameters as reference data to accurately determine (e.g., predict) measurements and other values of one or more physiological parameters.

When the computing system of the external device 24 and/or remote monitoring service 6 completes the above-described training process for the untrained decision tree-based model, the external device 24 and/or monitoring service 6 deploys the trained decision tree-based model to the IMD 16 to support the function of the medical device, for example, by accurately calculating pressure measurements and one or more derivatives of the volume within the left ventricle of the patient 14, while monitoring the patient's heart activity for abnormal rhythms to detect and/or correct. Thus, the decision tree-based model described above may be trained (offline) to utilize a resource-rich computing system and to reserve valuable resources at IMD 16.

Remote monitoring service 6 and/or external device 24 may utilize network communications with IMD 16 to receive actual sensor measurements in wireless communications for training a machine learning model to operate as an accurate estimation model. In response to receiving the actual sensor measurements, remote monitoring service 6 and/or external device 24 may, in turn, update (e.g., further train) an estimation model deployed at IMD 16 using the actual sensor measurements as reference data to more accurately estimate future sensor measurements. In one example, the pressure sensor 2 generates a signal encoding the actual pressure measurement and transmits the generated signal to the remote monitoring service 6 and/or the external device 24 via network communication. While IMD 16 may use the trained estimation model to determine an estimate of the same pressure measurement and then use the estimate in monitoring/correcting cardiac asynchrony, remote monitoring service 6 and/or external device 24 may push model updates to IMD 16.

In some examples, the processing circuitry of medical system 10 utilizes offline training for the decision tree-based model to customize the training process described above for patient 14 and/or IMD 16 and generate a personalized version of the trained model described above. In this way, the personalized decision tree-based model is configured to estimate left ventricular pressure specifically for the heart 12 of the patient 14 or for any similar patient. A personalized decision tree-based model may be deployed to customize IMD 16 (e.g., and some aspect of their physiology) for patient 14.

Many other applications are conceivable in view of this disclosure, including alternative/additional use cases of the systems, devices, and techniques described herein for determining the most likely future sensor measurements. According to one such use case, the estimation model may be configured to accurately predict which values are expected from a given sensor in the device, and the processing circuitry of medical system 10 may redirect the estimation model to a different purpose, such as surrogate prediction. Other applications show the advantage of having an estimation model (such as the decision tree-based model described above) in a medical device (such as IMD 16), with one exemplary use case being for enabling active alerts on behalf of patient 14.

Active alerting generally involves determining when human actions (e.g., interventions) are most likely to benefit the patient 14 in some way (e.g., improve patient health), and provides a number of benefits over reactive alerting, and depending on the physiology of the patient 14 (among other factors), may prevent abrupt and/or critical health events from occurring. Some embodiments implement active alerts by applying a decision tree-based model to predict (rather accurately) future sensor measurements expected at an upcoming point in time, and then outputting an active alert based on the future sensor measurements; while in other embodiments, the decision tree based model is integrated into a predictive model in which a plurality of factors determine when to output an active alert.

To demonstrate one or more benefits in the illustrative example, IMD 16 may incorporate a decision tree based model into the analysis for determining whether future sensor measurements justify an active alert, such as by outputting an alert intended to draw attention of a physician to the future sensor measurements, thereby prompting the physician to perform an action. For example, the alert may inform a physician that an emergency needs to be performed for an action that IMD 16 is not configured to perform. If patient 14 has a pacing device, some exemplary actions to be performed by a physician include increasing therapy/drug intake or calling another physician. It should be noted that other devices send active alerts for notifying the physician of patient 14, particularly in environments having significantly more resource capacity and capabilities than other embodiments of pacemakers and IMD 16. In one example, the processing circuitry of the computing system of the remote monitoring service 6 and/or the processing circuitry of the external device 24 outputs an alert for receipt by a physician via a local output device or a remote output device.

Fig. 2 is a conceptual diagram illustrating IMD 16 and leads 18, 20, 22 of medical device system 10 in greater detail. Leads 18, 20, 22 may be electrically coupled to therapy delivery circuitry, sensing circuitry, or other circuitry of IMD 16 via connector block 34. In some examples, the proximal ends of leads 18, 20, 22 include electrical contacts that are electrically coupled to corresponding electrical contacts within connector block 34. Additionally, in some examples, leads 18, 20, 22 are mechanically coupled to connector block 34 by means of a set screw, a connection pin, or another suitable mechanical coupling mechanism.

Each of the leads 18, 20, 22 includes an elongated insulated lead body that carries a plurality of conductors separated from one another by a tubular insulating sheath. In the illustrated example, bipolar electrode 40 and bipolar electrode 42 are located near the distal end of lead 18. In addition, bipolar electrode 44 and bipolar electrode 46 are located near the distal end of lead 20, and bipolar electrode 48 and bipolar electrode 50 are located near the distal end of lead 22. Electrode 40, electrode 44 and electrode 48 may take the form of ring electrodes and electrode 42, electrode 46 and electrode 50 may take the form of extendable helix tip electrodes telescopically mounted within insulated electrode head 52, insulated electrode head 54 and insulated electrode head 56, respectively. Each of electrodes 40, 42, 44, 46, 48, and 50 may be electrically coupled to a respective one of the conductors within the lead body of its associated lead 18, 20, 22, and thereby to a respective one of the electrical contacts on the proximal ends of leads 18, 20, and 22.

Electrode 40, electrode 42, electrode 44, electrode 46, electrode 48, and electrode 50 may sense electrical signals that accompany depolarization and repolarization of heart 12. Electrical signals are conducted to IMD 16 via respective leads 18, 20, 22. In some examples, IMD 16 also delivers pacing pulses to LV 32 via electrodes 44, 46 to cause depolarization of cardiac tissue of heart 12. In some examples, as shown in fig. 2, IMD 16 includes one or more housing electrodes, such as housing electrode 58, which may be integrally formed with an outer surface of hermetic housing 60 of IMD 16 or otherwise coupled to housing 60. In some examples, housing electrode 58 is defined by an uninsulated portion of an outward facing portion of housing 60 of IMD 16. Other divisions between the insulated and uninsulated portions of housing 60 may be used to define two or more housing electrodes. In some examples, the housing electrode 58 includes substantially all of the housing 60. Any of electrodes 40, 42, 44, 46, 48, and 50 may be used in combination with housing electrode 58 for monopolar sensing or stimulation delivery. As described in further detail with reference to fig. 3, the housing 60 may enclose a therapy delivery circuit that generates cardiac pacing pulses and defibrillation or cardioversion shocks, as well as a sensing circuit for monitoring the patient's heart rhythm.

In some examples, leads 18, 20, 22 may also include elongated electrodes 62, 64, 66, respectively, which may take the form of coils. IMD 16 may deliver defibrillation pulses to heart 12 via any combination of elongated electrodes 62, 64, 66 and housing electrode 58. Electrodes 58, 62, 64, 66 may also be used to deliver cardioversion pulses to heart 12. Electrodes 62, 64, 66 may be made of any suitable conductive material, such as, but not limited to, platinum alloys, or other materials known to be useful in implantable defibrillation electrodes.

The configuration of the medical system 10 shown in fig. 1 and 2 is one example and is not intended to be limiting. In other examples, the treatment system may include an extravascular electrode, such as a subcutaneous electrode, extra-cardiac electrode, and/or patch electrode, in addition to the transvenous leads 18, 20, and 22 electrodes shown in fig. 1. In addition, IMD 16 need not be implanted within patient 14. In examples where IMD 16 is not implanted within patient 14, IMD 16 may deliver defibrillation pulses, pacing pulses, and other therapies to heart 12 via percutaneous leads extending through the skin of patient 14 to various locations inside or outside of heart 12.

In other examples of a medical device system that provides electrical stimulation therapy to heart 12, the therapy system may include any suitable number of leads coupled to IMD 16, and each of the leads may extend to any location within or near heart 12. For example, the treatment system may include a dual lumen device instead of the three lumen device shown in fig. 1. In one example of a dual-chamber configuration, IMD 16 is electrically connected to a single lead that includes stimulation and sensing electrodes within LV 32 and sensing and/or stimulation electrodes within RA 26, as shown in fig. 3. In another example of a dual-cavity configuration, IMD 16 is connected to two leads that each extend into a respective one of RA 28 and LV 32.

In some examples, the medical device system includes one or more intracardiac pacing devices in lieu of or in addition to an IMD (e.g., IMD 16) coupled to leads extending to heart 12. The intracardiac pacing device may include therapy delivery and processing circuitry within a housing configured for implantation within one of the chambers of the heart 12. In such systems, one or more intracardiac pacing devices, which may include one or more leads, and/or a plurality of pacing devices of the IMD may communicate to coordinate sensing and pacing in various chambers of heart 12 to provide CRT in accordance with the techniques described herein. The processing circuitry and memory of one or more of the pacing device and/or another implantable or external medical device may provide functionality for controlling CRT delivery due to the processing circuitry and memory of IMD 16 herein.

As shown in fig. 2, IMD 16 is communicatively coupled to one or more sensors positioned to sense aspects of the patient's cardiac health. The sensors may be configured in locations for sensing cardiac activity, particularly various physiological parameters corresponding to the heart 12 (e.g., near the heart 12). Pressure sensor 2 is configured to generate signals (e.g., physiological parameter signals) to encode pressure data of heart 12 and avoid negative effects of employing pressure sensor 2 for these signals, IMD 16 uses the model to estimate the same pressure data; if the estimated pressure data meets minimum accuracy requirements, IMD 16 may employ the estimated pressure data in detection logic for monitoring cardiac abnormalities and/or in therapy circuitry to correct the cardiac abnormalities.

Although not shown in fig. 2, the system 10 may include one or more sensors, such as accelerometers or other mechanical sensors, operating as intermediate or indirect sensors of pressure data (e.g., left ventricular pressure). As an example, mechanical sensors may be located at different locations, such as within IMD 16, at endocardial or epicardial locations of heart 12 (e.g., locations 4A and 4B, on the free walls of the right and left ventricles proximate the mitral and tricuspid valves), or on the left ventricle and/or right ventricular apex. The sensor may be disposed on one or more of leads 18, 20, 22, on another lead not shown in fig. 2, or may be wirelessly coupled to IMD 16. It should be noted that the sensors need not be within IMD 10 or on the heart, and that instead, system 10 may employ at least one sensor of a body area network, which may be an in vivo wireless heart sensor network.

The mechanociceptors may each provide a first physiological parameter signal, such as an endocardial mechanoreceptive signal or an epicardial mechanoreceptive signal, that encodes a first physiological parameter. The system 10 may have one or more sensors positioned in alternative locations other than the endocardial or epicardial locations within the IMD 16 or referred to herein; for example, the mechanical sensor may be positioned in the vascular system of the patient 14, such as a vascular compartment (e.g., transvenous), rather than at the end of a cardiac lead (e.g., one or more of leads 18, 20, 22, or another lead not shown in fig. 2). When determining an estimate of the pulmonary or arterial pressure within the patient 14, the mechanical sensor may be configured as a stand-alone device.

Fig. 3 is a functional block diagram illustrating an exemplary configuration of IMD 16 of fig. 1 and 2. In the illustrated example, IMD 16 includes a memory 70, processing circuitry 80, sensing circuitry 82, one or more accelerometers 84, therapy delivery circuitry 86, telemetry circuitry 88, and a power source 90, one or more of which may be disposed within housing 60 of IMD 16.

In some examples, memory 70 includes computer readable instructions that, when executed by processing circuitry 80, cause IMD 16 and processing circuitry 80 to perform various functions attributed herein to IMD 16 and processing circuitry 80. The memory 70 may include any volatile memory, non-volatile memory, magnetic memory, optical memory, or dielectric memory, such as Random Access Memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), electrically Erasable Programmable ROM (EEPROM), flash memory, or any other digital media. In addition to sensed physiological parameters of patient 14 (e.g., EGM or ECG signals), one or more time intervals for fusion pacing therapy and biventricular pacing therapy timing of heart 12 may be stored by memory 70.

In general, the memory 70 may include various information data sets (e.g., database tables) and/or software components (e.g., software programs). As shown in the examples of fig. 3 and 4, the memory 70 may include model data 72 and patient data 74. In general, model data 72 refers to a machine learning model (e.g., a decision tree model or a neural network) that is applied to patient data 74.

The processing circuitry 80 may include one or more of the following: a microprocessor, a controller, a digital signal processing circuit (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or equivalent discrete or integrated logic circuit. In some examples, processing circuitry 80 may include a plurality of components (such as one or more microprocessors, one or more controllers, one or more DSPs, one or more ASICs, or any combinations of one or more FPGAs), as well as other discrete or integrated logic circuitry. The functions attributed to processing circuitry 80 herein may be embodied as software, firmware, hardware or any combination thereof. The processing circuitry 80 may be configured to determine the heart rate of the heart 12 based on the electrical activity sensed by the sensing circuitry 82, estimate left ventricular pressure data (e.g., LV pressure and/or its derivatives) from machine learning resources utilizing the model data 72 of the memory 70 and from various sensor data stored in the patient data 74 (e.g., which may be encoded into a structured dataset of physiological parameters), and control delivery CRT to the heart 12 based on the left ventricular pressure data by the therapy delivery circuitry 86. Further to this example, the processing circuitry 80 may be configured to cause the therapy delivery circuitry 86 to deliver electrical pulses.

Sensing circuitry 82 is configured to monitor signals from at least one of electrodes 40, 42, 44, 46, 48, 50, 58, 62, 64, or 66, for example, to monitor electrical activity of heart 12 via EGM signals. For example, sensing circuitry 82 may sense atrial events (e.g., P-waves) with electrodes 48, 50, 66 within RA 26, or LV 32 events (e.g., R-waves) with electrodes 44, 46, 64 within LV 32. In some examples, sensing circuitry 82 includes switching circuitry to select which of the available electrodes are used to sense electrical activity of heart 12. For example, the processing circuitry 80 may select electrodes for use as sense electrodes via switching circuitry within the sensing circuitry 82, such as by providing signals via a data/address bus. In some examples, the sense circuit 82 includes one or more sense channels, each of which may include an amplifier. In response to a signal from the processing circuit 80, the switching circuit of the sensing circuit 82 may couple the output from the selected electrode to one of the sense channels.

In some examples, one channel of sensing circuit 82 may include an R-wave amplifier that receives signals from electrodes 40 and 42 for pacing and sensing in RV 28 of heart 12. Another channel may include another R-wave amplifier that receives signals from electrode 44 and electrode 46 for pacing and sensing near LV 32 of heart 12. In some examples, the R-wave amplifier may take the form of an automatic gain control amplifier that provides an adjustable sensing threshold based on the R-wave amplitude of the measured heart rate.

Additionally, in some examples, one channel of sensing circuit 82 may include a P-wave amplifier that receives signals from electrodes 48 and 50 for pacing and sensing in RA 26 of heart 12. In some examples, the P-wave amplifier may take the form of an automatic gain control amplifier that provides an adjustable sensing threshold based on the P-wave amplitude of the measured heart rate. Examples of R-wave and P-wave amplifiers are described in U.S. patent No. 5,117,824 to Keimel et al, entitled "device for monitoring electrophysiological signals (APPARATUS FOR MONITORING ELECTRICAL PHYSIOLOGIC SIGNALS)", 6/2, which is incorporated herein by reference in its entirety. Other amplifiers may also be used. Further, in some examples, one or more of the sensing channels of sensing circuit 82 may be selectively coupled to housing electrode 58 or elongate electrode 62, electrode 64, or electrode 66, in addition to or in lieu of one or more of electrode 40, electrode 42, electrode 44, electrode 46, electrode 48, or electrode 50, e.g., for unipolar sensing of R-waves or P-waves in any of chambers 26, 28, or 32 of heart 12.

In some examples, the sensing circuit 82 includes a channel that includes an amplifier having a relatively wider passband than an R-wave or P-wave amplifier. The selected signal from the selected signal for coupling to this wideband amplifier may be provided to a multiplexer and thereafter converted by an analog to digital converter to a multi-bit digital signal for storage as an EGM in memory 70. In some examples, the storage of such EGMs in memory 70 may be under the control of direct memory access circuitry. Processing circuitry 80 may employ digital signal analysis techniques to characterize the digitized signals stored in memory 70 to detect and classify the patient's heart rhythm from the electrical signals. The processing circuitry 80 may detect and classify the heart rhythm of the patient 14 by employing any of a number of signal processing methods known in the art.

The signals generated by the sensing circuit 82 may include, for example: an RA event signal indicating detection of a P-wave via an electrode implanted within RA 26 (fig. 1); a LA event signal indicating detection of a P-wave via an electrode implanted within LA 33 (fig. 1); an RV event signal indicating the detection of R-waves via electrodes implanted within RV 28; or an LV event signal indicating the detection of R-waves via electrodes implanted within LV 32. In the example of medical system 10 shown in fig. 1 and 2, IMD 16 is not connected to electrodes implanted within LA 33. However, in other exemplary therapy systems, IMD 16 may be connected to electrodes implanted within LA 33 in order to sense electrical activity of LA 33.

In some examples, IMD 16 may include one or more additional sensors, such as accelerometer 84. In some examples, the accelerometer 84 may include one or more tri-axial accelerometers. As described above, accelerometer 84 may be an example of a mechanical sensor for determining estimated LV pressure data according to techniques described herein and may be located within the housing of IMD 16 or coupled to the IMD by one or more leads or wireless connections. The signals generated by the accelerometer 84 may be indicative of, for example, overall body movements of the patient 14, such as patient posture or activity level, and cardiac motion and vibration.

Therapy delivery circuitry 86 is electrically coupled to electrodes 40, 42, 44, 46, 48, 50, 58, 62, 64, and 66, for example, via conductors of respective leads 18, 20, 22, or, in the case of housing electrode 58, via electrical conductors disposed within housing 60 of IMD 16. The therapy delivery circuit 86 is configured to generate and deliver electrical stimulation therapy. For example, therapy delivery circuit 86 may deliver pacing stimulation to LV 32 (fig. 2) of heart 12 via at least two electrodes 44, 46 (fig. 2) according to the fusion pacing techniques described herein. As another example, therapy delivery circuit 86 may deliver pacing stimulation to RV 28 via at least two electrodes 40, 42 (fig. 2) and to LV 32 via at least two electrodes 44, 46 (fig. 2), e.g., according to the biventricular pacing techniques described herein.

In some examples, therapy delivery circuitry 86 is configured to deliver cardioversion or defibrillation shocks to heart 12. Pacing stimulation, cardioversion shocks, and defibrillation shocks may be in the form of stimulation pulses. In other examples, the therapy delivery circuit 86 may deliver one or more of these types of stimulation in the form of other signals, such as sine waves, square waves, or other substantially continuous time signals.

Treatment delivery circuit 86 may include a switching circuit, and processing circuit 80 may use the switching circuit to select (e.g., via a data/address bus) which of the available electrodes to use to deliver defibrillation or pacing pulses. The switching circuit may comprise a switching array, a switching matrix, a multiplexer, or any other type of switching device suitable for selectively coupling stimulation energy to selected electrodes. In other examples, processing circuitry 80 may select a subset of electrodes 40, 42, 44, 46, 48, 50, 58, 62, 64, and 66 with which to deliver stimulation to heart 12 without switching circuitry.

The processing circuitry 80 includes pacemaker timing and control circuitry 96, which may be embodied in hardware, firmware, software, or any combination thereof. Pacemaker timing and control circuitry 96 may include dedicated hardware circuitry (such as an ASIC) separate from other processing circuitry 80 components (such as a microprocessor), or software modules executed by components of processing circuitry 80 (e.g., a microprocessor or ASIC). Pacemaker timing and control circuitry 96 may help control the delivery of pacing pulses to heart 12.

In examples where IMD 16 delivers pacing pulses, pacemaker timing and control circuitry 96 may include a timer for determining that an atrial pace or sensed event has occurred (a _P/S Or more generally a) after determining that the selected a-V interval has elapsed. The timer of pacemaker timing and control circuit 96 may be configured to detect a previous atrial pace or sensed event (a _P/S ) And starts at that time. Upon expiration of a particular timer, processing circuitry 80 may control therapy delivery circuitry 86 to deliver pacing stimulation to heart 12 in accordance with a fusion or biventricular pacing configuration. For example, pacing timing and control circuitry 96 may generate trigger signals that trigger the output of pacing pulses via therapy delivery circuitry 86.

In examples where IMD 16 is configured to deliver other types of cardiac rhythm therapies in addition to fusion pacing and biventricular pacing, pacemaker timing and control circuitry 96 may also include programmable counters that control basic time intervals associated with DDD, VVI, DVI, VDD, AAI, DDI, DDDR, VVIR, DVIR, VDDR, AAIR, DDIR and other single-and dual-chamber pacing modes. In the foregoing pacing mode, "D" may indicate dual chambers, "V" may indicate ventricles, "I" may indicate inhibited pacing (e.g., no pacing), and "a" may indicate atria. The first letter in the pacing mode may indicate a chamber being paced, the second letter may indicate a chamber in which an electrical signal is sensed, and the third letter may indicate a chamber in which a response to sensing is provided.

In examples where IMD 16 is configured to deliver other types of cardiac rhythm therapy than CRT, the intervals defined by pacemaker timing and control circuitry 96 within processing circuitry 80 may include atrial and ventricular pacing escape intervals, refractory periods during which sensed P-waves and R-waves are ineffective to restart timing of the escape intervals, and pulse widths of pacing pulses. As another example, pacemaker timing and control circuit 96 may define blanking periods and provide signals from sensing circuit 82 to blank one or more channels, such as amplifiers, during and for a period of time after delivery of electrical stimulation to heart 12. The duration of these intervals may be determined by processing circuitry 80 in response to data stored in memory 70. In some examples, pacemaker timing and control circuit 96 of processing circuit 80 may also determine the amplitude of cardiac pacing pulses.

During a particular pacing mode, escape interval counters within pacemaker timing/control circuitry 96 of processing circuitry 80 may be reset upon sensing R-waves and P-waves. Treatment delivery circuitry 86 may include pacemaker output circuitry selectively coupled, for example by switching circuitry, to any combination of electrodes 40, 42, 44, 46, 48, 50, 58, 62, or 66 suitable for delivering bipolar or unipolar pacing pulses to one of the chambers of heart 12. Processing circuitry 80 may reset the escape interval counter upon generation of pacing pulses by therapy delivery circuitry 86 and thereby control the basic timing of cardiac pacing functions including fused cardiac resynchronization therapy.

Upon reset by sensed R-waves and P-waves, the count values present in the escape interval counter may be used by processing circuitry 80 to measure the duration of the R-R interval, P-P interval, P-R interval, and R-P interval, which are measurements that may be stored in memory 70. Processing circuitry 80 may use the count in the interval counter to detect a tachyarrhythmia event, such as a ventricular fibrillation event or a ventricular tachycardia event. When a threshold number of tachyarrhythmia events are detected, processing circuitry 80 may identify the presence of a tachyarrhythmia episode, such as a ventricular fibrillation episode, a ventricular tachycardia episode, or a non-sustained tachycardia (NST) episode. Examples of tachyarrhythmia episodes that may qualify for responsive therapy include ventricular fibrillation episodes or ventricular tachyarrhythmia episodes.

In some examples, processing circuitry 80 may operate as an interrupt-driven device and be responsive to interrupts from pacemaker timing and control circuitry 96, where the interrupts may correspond to sensed occurrences of P-waves and R-waves and generation of cardiac pacing pulses. Any necessary mathematical calculations are performed by the processing circuit 80, and any updating of values or intervals controlled by the pacemaker timing and control circuit 96 of the processing circuit 80 may occur after such an interrupt. A portion of memory 70 may be configured as a plurality of recirculation buffers capable of holding a series of measured intervals that may be analyzed by processing circuitry 80 in response to the occurrence of a pacing or sensing disruption to determine whether patient's heart 12 is currently exhibiting atrial or ventricular tachyarrhythmia.

In some examples, the arrhythmia detection method may include any suitable tachyarrhythmia detection algorithm. In one example, the processing circuitry 80 may utilize all or a subset OF the rule-based detection methods described in U.S. patent No. 5,545,182 entitled "priority rule-based methods and devices for diagnosing and treating ARRHYTHMIAS (PRIORITIZED RULE BASED METHOD AND APPARATUS FOR DIAGNOSIS AND TREATMENT OF arrhyhmias)" to Olson et al, month 13, 1996, or U.S. patent No. 5,755,736 entitled "priority rule-based methods and devices for diagnosing and treating ARRHYTHMIAS (PRIORITIZED RULE BASED METHOD AND APPARATUS FOR DIAGNOSIS AND TREATMENT OF arrhyhmias)" to gilnberg et al, month 26. U.S. patent No. 5,545,182 to Olson et al and U.S. patent No. 5,755,736 to gilnberg et al are incorporated herein by reference in their entirety. However, in other examples, processing circuitry 80 may also employ other arrhythmia detection methods.

If IMD 16 is configured to generate and deliver defibrillation shocks to heart 12, therapy delivery circuitry 86 may include high voltage charging circuitry and high voltage output circuitry. Where processing circuitry 80 determines that a cardioversion or defibrillation shock needs to be generated, processing circuitry 80 may employ an escape interval counter to control the timing of such cardioversion and defibrillation shocks and associated refractory periods. In response to detecting atrial or ventricular fibrillation or tachyarrhythmia requiring cardioversion pulses, processing circuitry 80 may activate cardioversion/defibrillation control circuitry (not shown), such as pacemaker timing and control circuitry 96, which may be hardware components of processing circuitry 80 and/or firmware or software modules executed by one or more hardware components of processing circuitry 80. The cardioversion/defibrillation control circuitry may initiate charging of the high voltage capacitor of the high voltage charging circuitry of the therapy delivery circuit 86 under the control of the high voltage charging control line.

The processing circuit 80 may monitor the voltage on the high voltage capacitor, for example, via a voltage charge and potential (VCAP) line. In response to the voltage on the high voltage capacitor reaching a predetermined value set by the processing circuit 80, the processing circuit 80 may generate a logic signal to terminate the charging. Thereafter, the timing of the delivery of defibrillation or cardioversion pulses by therapy delivery circuit 86 is controlled by the cardioversion/defibrillation control circuitry (not shown) of processing circuit 80. After delivering the defibrillation or tachycardia therapy, processing circuitry 80 may return therapy delivery circuitry 86 to cardiac pacing function and await the next sequential interruption due to pacing or the sensed occurrence of atrial or ventricular depolarizations.

The therapy delivery circuit 86 may deliver cardioversion or defibrillation shocks by means of an output circuit that determines whether monophasic or biphasic pulses are delivered, whether the housing electrode 58 is functioning as a cathode or an anode, and which electrodes are involved in the delivery of cardioversion or defibrillation pulses. Such functionality may be provided by one or more switches or switching circuits of the therapy delivery circuit 86.

Telemetry circuitry 88 includes any suitable hardware, firmware, software, or any combination thereof for communicating with another device, such as external device 24 (fig. 1). Telemetry circuitry 88 may receive downlink telemetry from external device 24 and transmit uplink telemetry to the external device by way of an antenna that may be internal and/or external under control of processing circuitry 80. Processing circuitry 80 may provide telemetry circuitry within telemetry circuitry 88 with data and control signals to be uplink to external device 24, for example, via an address/data bus. In some examples, telemetry circuitry 88 may provide the received data to processing circuitry 80 via a multiplexer.

In some examples, processing circuitry 80 may transmit atrial and ventricular heart signals (e.g., EGM signals) generated by atrial and ventricular sense amplifier circuitry within sensing circuitry 82 to external device 24. Other types of information, such as various intervals and delays for delivering CRTs, may also be transmitted to the external device 24. External device 24 may interrogate IMD 16 to receive cardiac signals. The processing circuitry 80 may store the cardiac signals within the memory 70 and retrieve the stored cardiac signals from the memory 70. The processing circuitry 80 may also generate and store a marker code indicative of the different heart attacks detected by the sensing circuitry 82 and transmit the marker code to the external device 24. An exemplary pacemaker with a marker channel capability is described in U.S. patent No. 4,374,382 entitled "marker channel telemetry system FOR medical devices (MARKER CHANNEL TELEMETRY SYSTEM FOR A MEDICAL DEVICE)" issued to Markowitz at month 2 and 15 of 1983, the entire contents of which are incorporated herein by reference.

Various components of IMD 16 are coupled to a power source 90, which may include a rechargeable or non-rechargeable battery. The non-rechargeable battery may be selected to last for years, while the rechargeable battery may be inductively charged from an external device, for example, on a daily or weekly basis. IMD 16 may utilize remote computing devices, such as external device 24 and/or remote monitoring service 6, to obtain additional resources (e.g., processing power) to, for example, perform training processes for generating trained versions of machine learning models to implement the parameter estimation techniques described herein.

Medical system 10 may be a computing system running in a single device or on a network of devices. Medical device IMD 16 is one exemplary device, but may also include other device types. IMD 16 includes a mechanical sensor configured to sense a first physiological parameter signal of a patient; and processing circuitry 80 configured to: determining one or more characteristics of the first physiological parameter signal; applying a machine learning model to the one or more features of the first physiological parameter signal; and determining an estimated value of the second physiological parameter based on the application of the machine learning model.

In one example, the medical system 10 may configure the mechanical sensor to sense a first physiological parameter signal that includes one or both of an endocardial mechanoreceptive signal or an epicardial mechanoreceptive signal encoding data (e.g., accelerometer data) of the first physiological parameter. Alternatively, the medical system 10 may configure the mechanical sensor to sense the transvenous mechanoreceptive signal as the first physiological parameter signal. Within IMD 16, the mechanical sensor may be configured to sense the first physiological parameter signal in an endocardial, epicardial, or thoracic body position of the patient.

One example of a second physiological parameter (pressure) may be estimated from a machine-learned model of model data 72. It should be noted that other parameters may be estimated using the same or similar techniques. In one example, medical system 10 may include processing circuitry, such as processing circuitry 80 of IMD 16, to determine an estimate of a parameter, such as a left ventricular pressure measurement, corresponding to the pressure of a chamber of a patient's heart. The model data 72 may include a readable and structured representation of a machine learning model such that the processing circuitry 80 may execute instructions for applying the model to one or more features extracted from the first physiological parameter signal. Based on the application, for example, the processing circuitry 80 is configured to generate an estimate of the left ventricular pressure measurement based on endocardial or epicardial mechanoreceptive signals.

After (or in lieu of) determining one or more estimates of the second physiological parameter (such as a left ventricular pressure measurement), the processing circuit 80 is further configured to calculate one or more derivatives of the particular left ventricular pressure measurement. If the particular left ventricular pressure measurement is based on the application of the machine learning model described above to endocardial or epicardial mechanoreceptive signals, then the one or more derivatives are also based on the endocardial or epicardial mechanoreceptive signals. The left ventricular pressure measurement and the estimation of any derivative value are based on one or more features extracted from endocardial or epicardial mechanoreceptive signals.

Generally, applying a machine learning model involves extracting one or more features from the first physiological parameter signal and then performing one or more (e.g., mathematical) functions using the one or more features as input data. In one example, the processing circuitry 80 extracts data for one or more features from the endocardial or epicardial mechanoreceptive signal or another exemplary first physiological parameter signal. Based on the processing circuitry 80 applying a machine learning model to such feature data, the processing circuitry 80 is configured to determine an estimate of the second physiological parameter (signal), such as the pressure data described above. In one example, the processing circuitry 80 is configured to calculate exemplary pressure data including an estimate of pressure measurements in a particular heart chamber of the patient. To calibrate the treatment response to cardiac asynchrony, processing circuitry 80 may calculate an estimated pressure measurement of the left ventricle. It should be noted that pressure data of another chamber may also be estimated, and these estimates may then be used to guide the treatment of the patient.

In one example, a machine learning model is applied to one or more features corresponding to a first physiological parameter encoded in an endocardial or epicardial mechanoreceptive signal. Accelerometer data may describe motion in a patient's heart and, as an exemplary parameter, may be encoded in endocardial or epicardial mechanoreceptive signals; these movements may be further processed into an estimate of a second physiological parameter, such as a pressure measurement (e.g., current or contemporaneous) of a chamber in the patient's heart.

IMD 16 of medical system 10 may configure processing circuitry 80 to use the estimated value of the second physiological parameter in a variety of applications, including monitoring/detection, therapy, and other functions. In an exemplary evaluation of patient data, an estimate of the (current) left ventricular pressure measurement may be used to determine whether the patient's heart activity is cardiac synchronized. The medical system may rely on such an assessment to determine whether to provide therapy to correct the abnormal heart rhythm. The processing circuitry 80 may execute logic configured to identify heart activity indicative of an abnormal heart rhythm. Based on one or more estimated pressure measurements of the left ventricle of the patient, the processing circuitry 80 may detect a condition of an abnormal heart rhythm. The processing circuitry 80 may be configured to apply a therapeutic response to the identified cardiac activity indicative of the abnormal heart rhythm, and the therapeutic response may be based on the estimated left ventricular pressure measurement as an exemplary second physiological parameter.

To further reduce false positives or false negatives, the processing circuitry 80 may perform an update of the machine learning model to achieve a higher level of accuracy. Such updating may be based on a comparison between an estimated value of the second physiological parameter (such as the left ventricular pressure measurement described above) and a reference value corresponding to the same or similar parameter. To illustrate an exemplary iteration of the training process performed by the computing system described above, the processing circuitry of medical system 10, in response to receiving an estimated left ventricular pressure measurement from the medical device, updates the machine learning model by comparing the estimated left ventricular pressure measurement for the patient's heart with reference pressure data and adjusting components of the machine learning model based on the comparison.

To train an initial version of the machine learning model to at least a minimum level of accuracy, the medical system 10 may include a computing system for running a training process on components of the model. In general, the training process evaluates a given machine learning model with respect to accuracy (e.g., an accuracy metric) by comparing an estimated value of a second physiological parameter to an actual (e.g., sensed or measured) parameter value and adjusting one or more model components (e.g., weights, thresholds, etc.) based on the comparison such that the adjusted machine learning model is configured to produce a modified estimated value that is more accurate than a previous estimated value. The machine learning model may be adjusted to reduce the difference (e.g., residual) between the estimated value of the same parameter and the corresponding reference value. The processing circuitry of medical system 10 may perform a training process on a corpus of cardiac activity data for a patient population or subset of patients and adjust the machine learning model for multiple iterations to increase the accuracy of each estimate. Finally, the processing circuitry of medical system 10 generates a training pattern of the machine learning model.

The above-described computing system (e.g., cloud computing environment) of medical system 10 may be communicatively coupled to a plurality of medical devices including IMD 16 via a network (e.g., a wireless network). After performing a training process on some training data (e.g., a corpus of cardiac activity data of a patient population or subset of patients) and generating a training pattern of a machine learning model to accurately determine an estimate of left ventricular pressure measurements (or another exemplary second physiological parameter), the processing circuitry of the computing system described above may be configured to deploy a (trained) machine learning model for implementation in IMD 16 or another medical device configured to monitor cardiac activity of the patient and provide a therapeutic response, if possible.

Examples of mechanical sensors that sense patient activity include accelerometers (e.g., tri-axial accelerometers), gyroscopes, thermometers, torque transducers, and the like. There are a number of methods for converting the encoded patient activity into one or more physiological parameters, each of which may be a quality (e.g., high activity, low activity, etc.) or a quantity (e.g., number of activity minutes or fraction of activity minutes (e.g., 10 seconds blocks)) representing some aspect of the patient's physiology. The various metric (e.g., accuracy metric) levels enable standardized measurements to be made on each sample of physiological parameter data (e.g., a timestamp), and enable differentiation between multiple samples of physiological parameter data (e.g., a timestamp or longer period and/or a patient).

In some examples, processing circuitry 80 executes detection logic to monitor for abnormalities in cardiac rhythm and/or diseases known as cardiac events, including arrhythmias, which may cause a patient's health to drop or otherwise negatively affect the patient's heart. The detection logic may be configured to determine whether the cardiac EGM data or ECG data is indicative of such a heart attack, e.g., based on one or more criteria. As described herein, the processing circuitry 80 may rely on alternative data sources for information describing one aspect of the heart activity of the patient 14. Instead of a particular sensor and its measurements, the processing circuitry 80 may utilize a model (e.g., a machine learning model) to estimate such measurements and then replace the sensor data in the detection analysis. The sensor data may be generated by one or more of the sensors 62 directly sensing data from within the body of the patient 14, particularly the heart of the patient 14.

In some examples, processing circuitry 80 incorporates a decision tree model into the detection logic and then applies the decision tree model to estimate left ventricular pressure data from the input feature data as an improvement over using a pressure sensor to directly sense left ventricular pressure from the heart chamber. In some examples, processing circuitry 80 may use a decision tree model to calculate the derivative of left ventricular pressure data. Instead of a conventional sensor device (e.g., a pressure sensor) that directly senses left ventricular pressure and its derivatives, the processing circuit 80 utilizes endocardial or epicardial mechanoreceptive signals (e.g., accelerometers) for patient activity data for calculating an estimate of left ventricular pressure (e.g., at the same timestamp). There are multiple mechanisms for measuring patient activity at a particular time, such as the 10 second 23 count method, which checks whether the integrated count of the front-side (z-axis) accelerometer reaches the 23 count threshold within each consecutive 10 second window. Processing circuitry 80 of IMD 16 may apply the 10 second 23 count method to an integral count over a predetermined period of time (e.g., minutes or hours).

Fig. 4 is a block diagram showing an exemplary configuration of components of the external device 24. In the example of fig. 4, the external device 24 includes processing circuitry 180, communication circuitry 182, storage 184, and a user interface 186.

The processing circuitry 180 may include one or more processors configured to implement functions and/or processing instructions for execution within the external device 24. For example, the processing circuitry 180 may be capable of processing instructions stored in the storage 184. The processing circuit 180 may comprise, for example, a microprocessor, DSP, ASIC, FPGA, or equivalent discrete or integrated logic circuit, or a combination of any of the foregoing devices or circuits. Thus, the processing circuitry 180 may comprise any suitable structure, whether hardware, software, firmware, or any combination thereof, to perform the functions attributed to the processing circuitry 180 herein.

Communication circuitry 182 may include any suitable hardware, firmware, software, or any combination thereof for communicating with another device, such as IMD 16. Communication circuitry 182 may receive downlink telemetry from IMD 16 or another device and transmit uplink telemetry to the IMD or another device under control of processing circuitry 180. The communication circuit 182 may be configured to transmit and/or receive signals via inductive coupling, electromagnetic coupling, NFC, RF communication, bluetooth, wiFi, or other proprietary or non-proprietary wireless communication schemes. Communication circuitry 182 may also be configured to communicate with devices other than IMD 16 via any of various forms of wired and/or wireless communication and/or network protocols.

The storage 184 may be configured to store information within the external device 24 during operation. The storage 184 may include a computer-readable storage medium or a computer-readable storage device. In some examples, the storage 184 includes one or more of short term memory or long term memory. The storage 184 may include, for example, RAM, DRAM, SRAM, magnetic disk, optical disk, flash memory, or various forms of EPROM or EEPROM. In some examples, the storage 184 is to store data indicative of instructions for execution by the processing circuitry 180. The storage 184 may be used by software or applications running on the external device 24 to temporarily store information during program execution.

Data exchanged between external device 24 and IMD 16 may include operating parameters. External device 24 may transmit data including computer readable instructions that, when implemented by IMD 16, may control IMD 16 to alter one or more operating parameters and/or derive the collected data. For example, processing circuitry 180 may transmit instructions to IMD 16 requesting IMD 16 to export the collected data (e.g., asystole episode data) to external device 24. External device 24 may then receive the collected data from IMD 16 and store the collected data in storage 184, e.g., as part of model data 72 and/or patient data 74. The data received by external device 24 from IMD 16 may include patient data 74, and similar to fig. 3, patient data 74 includes patient physiological parameter data, episode data (e.g., cardiac EGM), patient activity, and other sensor data. The processing circuit 180 may implement any of the techniques described herein to use the model data 72 as an estimation model for predicting values of current or future sensor measurements. As described, examples of sensor measurements may correspond to sensed patient heart activity. Similar to processing circuitry 80 of IMD 16, processing circuitry 180 may use the estimation model to analyze patient data 74 from IMD 16 and determine an estimate of the measurement of pressure sensor 2 at one or more points in time, e.g., to determine whether patient 14 is experiencing a change in health based on one or more criteria.

A user, such as a clinician or patient 14, may interact with the external device 24 through the user interface 186. User interface 186 includes a display (not shown), such as a Liquid Crystal Display (LCD) or a Light Emitting Diode (LED) display or other type of screen, that processing circuitry 180 may use to present information related to IMD 16, such as patient data 74 as described herein. Additionally, the user interface 186 may include an input mechanism configured to receive input from a user. The input mechanisms may include any one or more of, for example, buttons, a keypad (e.g., an alphanumeric keypad), a peripheral pointing device, a touch screen, or another input mechanism that allows a user to navigate through a user interface presented by the processing circuitry 180 of the external device 24 and provide input. In other examples, the user interface 186 also includes audio circuitry for providing audible notifications, instructions, or other sounds to the user, receiving voice commands from the user, or both.

Fig. 5 is a block diagram illustrating an exemplary system including an access point 190, a network 192, an external computing device (such as a server 194), and one or more other computing devices 199A-199N (collectively, "computing devices 199") that may be coupled to IMD 16 and external device 24 via network 192, in accordance with one or more techniques described herein. In this example, IMD 16 may use communication circuitry 154 to communicate with external device 24 via a first wireless connection and to communicate with access point 190 via a second wireless connection. In the example of fig. 5, access point 190, external device 24, server 194, and computing device 199 are interconnected and can communicate with each other through network 192.

Access point 190 may comprise a device connected to network 192 via any of a variety of connections, such as telephone dialing, digital Subscriber Line (DSL), or cable modem connections. In other examples, access point 190 may be coupled to network 192 through a different form of connection, including a wired connection or a wireless connection. In some examples, the access point 190 may be a user device, such as a tablet or smart phone, that may be co-located with the patient. IMD 16 may be configured to transmit data to access point 190, such as patient data 74 describing heart activity of patient 14 over a predetermined period of time and/or indications of patient health changes. The access point 190 may then transmit the retrieved data to the server 194 via the network 192.

As described herein, medical devices other than IMD 16 may generate data by directly sensing patient cardiac activity from locations within the patient's body (particularly within the patient's heart). Instead, IMD 16 and similar medical devices forego direct sensing and implement machine learning models and techniques for estimating left ventricular pressure as an alternative to data generated by direct sensing. Although the estimated left ventricular pressure data may depend on characteristic data from one or more sensors, IMD 16 does not utilize typical or conventional sensors to directly sense left ventricular pressure. Any sensor data used in the estimation is generated from non-invasive and/or passive components.

In some cases, server 194 may be configured to provide a secure storage site for data that has been collected from IMD 16 and/or external device 24. In some cases, server 194 may assemble data in a web page or other document via computing device 199 for viewing by trained professionals, such as clinicians. One or more aspects of the illustrated system of fig. 5 may be used with the force of meitonThe network provides general network technology and functions that are similar.

In some examples, one or more of computing devices 199 may be a tablet or other smart device located with a clinician through which the clinician may program to receive alerts and/or interrogate IMD 16. For example, a clinician may access patient data including estimated left ventricular pressure measurements and/or indications of patient health collected by IMD 16 via computing device 199, such as when patient 14 is between clinician visits, to check the status of the medical condition. In some examples, a clinician may input instructions for medical intervention of patient 14 into an application executed by computing device 199, such as based on a status of a patient condition determined by IMD 16, external device 24, server 194, or any combination thereof, or based on other patient data known to the clinician. Subsequently, the device 199 may transmit instructions for medical intervention to another one of the computing devices 199 located with the patient 14 or a caretaker of the patient 14. For example, such instructions for medical intervention may include instructions to change the dosage, timing, or selection of a drug, instructions to schedule a clinician visit, or instructions to seek medical attention. In further examples, computing device 199 may generate alerts to patient 14 based on the status of the medical condition of patient 14, which may enable patient 14 to actively seek medical attention prior to receiving instructions for medical intervention. In this way, the patient 14 may be authorized to take action as needed to address his medical condition, which may help improve the clinical outcome of the patient 14.

In the example illustrated by fig. 5, server 194 includes, for example, a storage device 196 and a processing circuit 198 for storing data retrieved from IMD 16. Although not shown in fig. 5, computing device 199 may similarly include a storage device and processing circuitry. The processing circuitry 198 may include one or more processors configured to implement functions and/or processing instructions for execution within the server 194. For example, the processing circuitry 198 may be capable of processing instructions stored in the storage 196. The processing circuit 198 may comprise, for example, a microprocessor, DSP, ASIC, FPGA, or equivalent discrete or integrated logic circuit, or a combination of any of the preceding devices or circuits. Thus, the processing circuitry 198 may comprise any suitable structure, whether hardware, software, firmware, or any combination thereof, to perform the functions attributed to the processing circuitry 198 herein. Processing circuitry 198 of server 194 and/or processing circuitry of computing device 199 may implement any of the techniques described herein to analyze various information received from IMD 16 for the patient, such as (further) evaluating a left ventricular pressure estimate determined by a model at IMD 16, (further) training a model used by IMD 16 to determine a left ventricular pressure estimate, (further) performing functions that support a remote monitoring service, and determining whether the health status of the patient has changed.

Storage 196 may comprise a computer-readable storage medium or a computer-readable storage. In some examples, storage 196 includes one or more of short term memory or long term memory. Storage 196 may comprise, for example, RAM, DRAM, SRAM, magnetic disk, optical disk, flash memory, or various forms of EPROM or EEPROM. In some examples, storage 196 is used to store data indicative of instructions for execution by processing circuitry 198.

The following description relates to the flowcharts of fig. 6A to 6C and fig. 7A to 7C, which respectively show examples of the first machine learning technique and the second machine learning technique. As described herein, the first machine learning technique enables non-invasive care from a medical device. Fig. 6A-6C further describe the first machine learning technique as a pipeline (e.g., an automated processing pipeline) for long-term tracking of cardiac function via mechanical feedback classification using machine learning. When implemented in a medical device (e.g., an implantable cardiac monitoring device), the pipeline enables additional functionality that supports cardiac monitoring applications. In some examples, the patient also benefits from non-invasive care from the medical device.

Fig. 6A is a flow diagram illustrating a first phase of a first machine learning technique in accordance with one or more examples of the present disclosure.

The first stage generally includes a training process for a machine learning model. In fig. 6A, a first stage is visualized as a process flow of generating an initial estimation model to implement additional functions of the medical device. Each additional function is beneficial to the patient and the medical device in at least one aspect as described herein. In view of the availability of training data sets and the general nature of machine learning training, the present disclosure contemplates a wide variety of additional functions; as one example of an additional function of normal device operation, the medical device may use an estimation model to estimate left ventricular pressure data as an alternative to directly sensing pressure in the left ventricle of the patient's heart. While estimating left ventricular pressure data is one example, a number of additional functions are envisioned from the present disclosure.

Fig. 6A illustrates the training process of the machine learning model described above to enable accurate estimation of left ventricular pressure measurements. It should be noted that other additional functionality contemplated by the present disclosure may be generated via the first machine learning technique. According to a first machine learning technique, a training process utilizes recorded data consisting of: a representation of patient cardiac activity (e.g., cardiac EGM) (210), cardiac feedback (e.g., mechanical feedback) from medical device components in endocardial, epicardial, and/or thoracic body locations (212), and reference signals (e.g., estimated signals) encoding actual measurements from one or more sensors (e.g., pressure sensor 2 of fig. 1 and 2).

In some examples of the first machine learning technique, the training process performs beat-to-beat segmentation (216) of the recorded data, wherein for each heartbeat, at least one actual sensor measurement (e.g., corresponding to heart 12) is compared to at least one corresponding estimate generated by a machine learning model (i.e., an estimation model), and based on the comparison, at least one model component (e.g., a parametric weight) is adjusted to improve accuracy of the model. For the first iteration of the training process, the machine learning model may be untrained or insufficiently trained to estimate the value of the physiological parameter of the patient. Over multiple iterations, the machine learning model may be fully trained and thus adapted to monitor the patient's cardiac health for abnormalities (e.g., cardiac asynchrony) and, if possible, provide a therapeutic response to correct/treat the abnormalities.

The training process determines adjustments to be made based on whether the resulting model generates a more accurate estimate of at least one corresponding estimate. Processing circuitry as described herein may use a machine learning model to predict sensor measurements (e.g., pressure measurements of the left ventricle) by calculating estimates from computational algorithms or mathematical functions. The computing algorithm or mathematical function may include input variables corresponding to various characteristic values. The processing circuitry may execute the software code to perform feature extraction (218) on the training data set and determine various feature values for the computational algorithm or mathematical function. Based on the difference between the estimated value and the corresponding actual sensor measurement (e.g., the sensor measurement of the recorded reference data), the model is adjusted by modifying the calculation algorithm or mathematical function to reduce the difference. Thus, the adjusted model may be used to calculate new estimates that more closely correspond to actual sensor measurements.

After multiple adjustments, the training process may generate an initial estimation model (220) for final deployment to the medical device. In one example, the training process employs a corpus of cardiac activity data for a population of patients and generates an initial estimation model as a generic model that is applicable to most, if not all, patients. In other examples, the training process takes cardiac activity data for a subset of patients or a particular patient (e.g., patient 14 of fig. 1) and generates an initial estimation model that is a personalized model.

To build the personalized model offline according to the first machine learning technique, the external computing device performs a training process and divides the training data into patient groups and/or individual patients. Patients may be grouped in a variety of ways, in one example, by patient demographics). The training process primarily coordinates the training of each personalized model using training data for similar patients in a particular group and/or for a particular patient.

As described herein, autonomous analysis is performed on the signal, wherein beats are segmented and predetermined features are extracted on a beat-by-beat basis. The initial record allows personalizing the model of the respective patient of the device to be received. External data sources may be used directly instead of patient records to train a more generalized and robust model. A combination of these two methods may be used as a plurality of machine learning applications for which a personalized model may be attached to a previously generated generic model.

Once trained, these personalized models can estimate cardiac functional pressure or selected measurements on a beat-to-beat basis by using implanted mechanical sensors for intermediate measurement data for identifying CRT non-responders and heart failure regression. This eliminates the need for direct sensing of indicative features. Benefits are seen especially when sensing pressure with the heart chamber, as pressure acquisition via the catheter is hindered by overgrowth, catheter leakage and sensor drift.

Fig. 6B is a flow diagram illustrating a second phase of the first machine learning technique of fig. 6A in accordance with one or more examples of the present disclosure.

Generally, the second stage refers to the process of fine-tuning the estimation model described above to prepare for deployment in the medical device, which is the third stage of the first machine learning technique. As shown in fig. 6A, the first stage represents a first stage of a first machine learning technique in which the estimation model undergoes the training process described above. In fig. 6B, the second stage is visualized as a process flow that determines whether the initial estimation model is sufficiently accurate within a predetermined degree of error, and outputs the model as a final estimation model if the initial estimation model meets the necessary level of accuracy.

As described herein, the training process of the machine learning based estimation model requires a training data dataset from reference data provided by the medical device and/or an external data source. In addition to the various reference data sources, many factors guide the flow of the second stage of processing. As an exemplary factor, different machine learning concepts may be used to generate different/alternative versions of the estimation model described above. FIG. 6B illustrates a machine learning method (222), a clustering algorithm (224), and a regression technique (226) as representations of different machine learning concepts. The accuracy of each resulting estimation model is evaluated and in a second stage, the best estimation model is selected (228). In some examples, in the second phase, the fine tuning process is to adjust the estimation model components (e.g., parameters such as node weights of the neural network) to improve the current metric score and/or further training.

Fig. 6B illustrates testing/evaluation of an initial estimation model that takes into account one or more metrics (e.g., for evaluating accuracy). The second stage of the fine tuning process accepts as input the initial estimation model and after evaluating the different embodiments of the model, the process flow selects the embodiments as the final estimation model. As shown, the second stage of the process flow selects the embodiment with the highest performance estimation model that is automatically compiled (230) and then deployed (e.g., installed) as an update to the medical device (232). In some examples, the update adds new functionality and/or replaces/modifies (e.g., as an improvement) existing functionality.

Fig. 6C is a flow diagram illustrating a third stage of the machine learning technique of fig. 6A in accordance with one or more examples of the present disclosure.

After implementing the compiled model, the device can use mechanociceptive feedback (242) to estimate characteristics such as absolute maximum pressure or maximum pressure variation (244) and can be used as a reference in adjusting therapy, for example, by improving pacing ramp settings (246). Similar to the described monitoring application, pacing control functions and functional studies may be trained to respond to estimated pressure values by adjusting pacing settings or heart rate to improve cardiac function.

Fig. 7A-7C illustrate cloud-based generation of additional medical device functionality using the second machine learning techniques described herein. The medical device operates within the patient's local environment, while the cloud computing device (e.g., cloud computing environment) operates at a remote location and over a data network; however, the medical device and the cloud's networked devices cooperate to form a medical system as also described herein, e.g., where a communication channel is generated between the two devices and a device-supported data analysis pipeline is enabled.

Fig. 7A is a flowchart illustrating operations 310, which are a first operation of a medical system running on a cloud computing environment, according to one or more examples of the present disclosure. The medical system enables cloud-based functional generation to be incorporated (e.g., programmed) into hardware/software of a medical device for cardiac applications. The additional functionality is based on the second machine learning technique described herein.

Operation 310 represents an automated process running in a cloud computing environment as described herein. Similar to the first machine learning technique of fig. 6A-6C, the second machine learning technique of fig. 7A-7C includes a training process followed by testing and then deployment as a cardiac monitoring application (e.g., remote monitoring service 6 of fig. 1) for a medical device. The second machine learning technique enables a device-supported data analysis pipeline via a communication channel between each medical device and the cardiac monitoring application.

The medical device performs one or more functions to facilitate its user, who may be a patient who is receiving medical care in some form. Computing devices in the cloud computing environment operate a remote monitoring service for these medical devices and are configured to analyze cardiac activity data received from the patient's medical devices. Patients receive medical care and device support from computing devices (e.g., data centers) in a cloud computing environment via their medical devices. According to operation 310, the computing device contains/manages data submitted/acquired from the patient's medical device and other implanted medical devices.

The computing device may be configured as a server for performing an automated process (1) for analyzing cardiac activity data and subsequently using the process results to train/update an estimation model of left ventricular pressure data. For example, a computing device may receive data for analysis via an automated processing pipeline as described herein. In furtherance of operation 310, a computing device may perform a beat-to-beat analysis (e.g., regardless of data type) of a recorded (e.g., segmented) cardiac ECG signal and extract one or more features on a beat-to-beat basis. The computing device may program modifications into the pipeline, for example by adding/removing features to be extracted and subsequently analyzed in a beat-to-beat analysis.

The computing device may configure the estimation model to provide an estimate of the measurement related to the sensed aspect of the patient's heart health. For example, a pressure sensor within a device located near the patient's heart may directly sense left ventricular pressure data; alternatively, the device may use an estimation model to predict the pressure data most likely to be sensed by the device. As part of the beat-to-beat analysis, the computing device receives reference data including sensed pressure measurements (e.g., or equivalent sensor data) generated by directly sensing pressure in the left ventricle. The computing device compares the reference data with an estimate of the same pressure measurement, and based on the comparison, the computing device modifies one or more estimation model components to generate a more accurate pressure estimate.

Each sensor measurement as described herein may be configured to quantify some aspect of patient activity such that combining one or more measurements may provide a comprehensive view or assessment of patient health. The estimation model may be configured with any number of model components (e.g., weights, formulas, methods, etc.) for calculating an estimate of the actual pressure within the left ventricle. After calculating the pressure value, processing circuitry of the computing device compares the estimated pressure value to a baseline value. In some examples, the baseline value may be a reference value generated by a pressure sensor that directly senses left ventricular pressure data. Over time, the reference pressure values may form a range, and the range corresponds to a baseline representing a typical patient health or normal heart health of an average patient. Deviations from the range (which deviations are statistically significant or exceed a predetermined threshold) are indicative of patient health status changes. In other examples, the baseline value may be predetermined or, alternatively, calculated using a formula and an alternative to the pressure sensor. In some examples, the baseline value is representative of the normal health condition of the particular patient, and any deviation from the baseline value should be assessed. The baseline value may represent a boundary of the patient because the baseline value is the highest pressure value/level while still indicating that the patient's health condition is not decreasing; any deviation from the baseline value may be indicative of an acute change or decrease in the patient's health condition.

In view of the widely available features, computing devices may execute various machine learning algorithms to generate suitable estimation models for determining data for use in one or more cloud-based applications/services. The computing device may evaluate the reference pressure measurements to further train the estimation model. There are many learning algorithms for evaluating a model to determine whether the model is sufficiently trained. Supervised (e.g., linear regression) and unsupervised learning (e.g., clustering) methods may generate differences between reference measurements and corresponding model estimates, and then plot the differences over time to visually see changes and trends. The linear regression line may be displayed to show a general trend, and based on the fitness (e.g., fitness) of the regression line to the reference line, the computing device may determine the level of accuracy of the predictions so far generated by the model. If the level of accuracy indicates an accurate and thus well-trained estimation model, e.g., in terms of specificity and/or sensitivity, the computing device may generate different embodiments (2) of the same estimation model, wherein each embodiment is a state detection function or an estimation function.

The computing device may incorporate a state detection function or an estimation function in a plurality of cardiac applications. The computing device may use any criteria to determine which alternative cardiac application function uses the trained estimation model as an alternative to the sensor data (3). Cloud-based applications/services may employ state detection functions or estimation functions to perform patient monitoring functions (e.g., use a machine learning model to estimate cardiac performance via mechanical feedback), pacing control functions (e.g., train and then use a machine learning model to determine effective pacing settings to maximize CRT response), research functions (e.g., add surveys of factors affecting heart failure patients, such as step tracking and fall detection for such patients, etc.).

Fig. 7B is a flowchart illustrating operations 320, which are a second operation of the medical system of fig. 7A, according to one or more examples of the present disclosure.

Operation 320 represents a device-side (e.g., client-side) application process of the medical system formed by operation 310, which is a cloud-based (e.g., server-side) application process running in a cloud computing environment. The two application processes combine to form a pipeline for adding functionality to the medical device, which is shown in fig. 7A-7B.

The physician (4) provides medical care to the patient (5). Fig. 7B illustrates physician (4) requesting externally provided services from the cloud computing environment performing operation 310. For example, a physician may connect a medical device to a cloud-based service, and when a patient is undergoing a medical examination, the physician may use a list of pre-made device features that may aid in remote monitoring and/or processing. Via the medical device, the physician downloads one or more models and feature data for compilation into the medical device. Additionally, the physician may instruct the medical device to upload patient data for storage in the cloud computing environment. In some cases, patient data may be used to further enhance cloud-based services, for example, by enabling estimation of one or more parameters (e.g., left ventricular pressure data) for sensed patient activity without directly sensing the one or more parameters.

Operation 320 of the second machine learning technique may add new or modified functionality to the medical device, which may or may not be implemented in a manner similar to fig. 6A-6C. Fig. 7B depicts the medical device as a pacemaker and further depicts updating the pacemaker with new/modified functionality, for example by programming one or more added functionality into the hardware/software of the pacemaker. Similar to fig. 6A-6C, new/modified pacemaker functions configured into the pacemakers described herein may include patient monitoring functions, pacing control functions, and research functions. The additional pacemaker function may extend (e.g., enhance) patient monitoring functions to include therapy delivery. If the pacemaker is configured to provide treatment or some other form of therapy to the patient, the pacemaker may reassign those treatments and/or adjust device treatment methods in order to improve patient treatment response.

The order and flow of operations shown in fig. 6 and 7 are examples. In other examples according to the present disclosure, more or fewer thresholds may be considered. Further, in some examples, as directed by the user, the processing circuitry may or may not perform the methods of fig. 6 and 7 or any of the techniques described herein, e.g., via external device 24 or computing device 100. For example, a patient, clinician, or other user may turn on or off functionality for identifying a change in patient health condition (e.g., using Wi-Fi or cellular services) or locally (e.g., using an application provided on the patient's cellular telephone or using a medical device programmer).

The techniques described in this disclosure may be implemented, at least in part, in hardware, software, firmware, or any combination thereof. For example, aspects of the techniques may be implemented in one or more microprocessors, DSP, ASIC, FPGA, or any other equivalent integrated or discrete logic QRS circuit, as well as any combination of such components, such components being embodied in an external device (such as a physician or patient programmer, simulator, or other device). The terms "processor" and "processing circuit" may generally refer to any of the foregoing logic circuits alone or in combination with other logic circuits or any other equivalent circuits alone or in combination with other digital or analog circuits.

For various aspects implemented in software, at least some of the functionality attributed to the systems and devices described in this disclosure may be embodied as instructions on a computer-readable storage medium, such as RAM, DRAM, SRAM, magnetic disk, optical disk, flash memory, or various forms of EPROM or EEPROM. The instructions may be executed to support one or more aspects of the functionality described in this disclosure.

In addition, in some aspects, the functionality described herein may be provided within dedicated hardware and/or software modules. Depiction of different features as modules or units is intended to highlight different functional aspects and does not necessarily imply that such modules or units must be realized by separate hardware or software components. Rather, functionality associated with one or more modules or units may be performed by separate hardware or software components or integrated within common or separate hardware or software components. In addition, the present technology may be fully implemented in one or more circuits or logic elements. The techniques of this disclosure may be implemented in various apparatuses or devices including an IMD, an external programmer, a combination of an IMD and an external programmer, an Integrated Circuit (IC) or a set of ICs and/or discrete circuits residing in an IMD and/or an external programmer.

Various examples have been described. These and other examples are within the scope of the following claims.

Example 1: a medical system, the medical system comprising: a mechanical sensor configured to sense a first physiological parameter signal of a patient; and processing circuitry configured to: determining one or more characteristics of the first physiological parameter signal; applying a machine learning model to the one or more features of the first physiological parameter signal; and determining an estimated value of a second physiological parameter based on the application of the machine learning model.

Example 2: the medical system of embodiment 1, wherein the mechanical sensor is configured to sense a first physiological parameter signal comprising at least one of an endocardial mechanoreceptive signal or an epicardial mechanoreceptive signal encoding cardiac motion data.

Example 3: the medical system of any one of embodiments 1 and 2, wherein the mechanical sensor comprises an accelerometer.

Example 4: the medical system of any one of embodiments 1-3, wherein to determine the estimated value of the second physiological parameter, the processing circuit is further configured to generate an estimate of a left ventricular pressure measurement based on the endocardial mechanoreceptive signal or the epicardial mechanoreceptive signal.

Example 5: the medical system of any one of embodiments 1-4, wherein to determine the estimated value of the second physiological parameter, the processing circuit is further configured to calculate one or more derivatives of left ventricular pressure measurements based on the endocardial mechanoreceptive signal or the epicardial mechanoreceptive signal.

Example 6: the medical system of any of embodiments 1-5, wherein to apply a machine learning model to the one or more features of the first physiological parameter signal, the processing circuit is further configured to apply the machine learning model to accelerometer data and determine an estimate of pressure measurements in a heart chamber based on the application of the machine learning model.

Example 7: the medical system of any one of embodiments 1-6, wherein to determine the estimated value of the second physiological parameter, the processing circuit is further configured to determine estimated pressure data.

Example 8: the medical system of any one of embodiments 1-7, wherein the processing circuit is further configured to update the machine learning model based on a comparison of the estimated value to a reference value corresponding to the second physiological parameter.

Example 9: the medical system of any one of embodiments 1-8, wherein the processing circuit is further configured to generate the machine learning model by performing a training process on a corpus of first physiological parameter data and second physiological parameter data of a patient population or a subset of patients.

Example 10: the medical system of any one of embodiments 1-9, wherein the processing circuit is further configured to use the estimated value of the second physiological parameter in an assessment of cardiac synchrony.

Example 11: the medical system of any one of embodiments 1-10, wherein the processing circuit is further configured to provide therapy to correct cardiac asynchrony using the estimated value of the second physiological parameter.

Example 12: a medical system comprising a communication circuit coupled to a medical device through a network; and processing circuitry configured to: performing a training process on a corpus of cardiac activity data of a patient population or a subset of patients; generating a machine learning model based on the training process, wherein the machine learning model is configured to determine an estimate of a second physiological parameter using the first physiological parameter signal; and deploying the machine learning model for use in a medical device configured to monitor heart activity of a patient.

Example 13: the medical system of embodiment 12, wherein the processing circuit is further configured to: in response to receiving an estimated left ventricular pressure measurement from the medical device, the machine learning model is updated by comparing the estimated left ventricular pressure measurement for the patient's heart to reference pressure data and adjusting components of the machine learning model based on the comparison.

Example 14: a method comprising determining one or more characteristics of a first physiological parameter signal sensed by a mechanical sensor; applying a machine learning model to the one or more features of the first physiological parameter signal; and determining an estimated value of a second physiological parameter based on the application of the machine learning model.

Example 15: the method of embodiment 14, wherein the mechanical sensor comprises an accelerometer.

Example 16: the method of any one of embodiments 14 and 15, wherein determining an estimated value of the second physiological parameter further comprises determining estimated pressure data.

Example 17: the method according to any one of embodiments 14-16, further comprising using the estimated value of the second physiological parameter in an assessment of cardiac synchrony.

Example 18: the method according to any one of embodiments 14-17, further comprising using the estimated value of the second physiological parameter to provide a treatment to correct cardiac asynchrony.

Example 19: the method of any of embodiments 14-18, further comprising identifying cardiac activity indicative of an abnormal heart rhythm based on the estimated value of the second physiological parameter.

Example 20: the method of any of embodiments 14-19, further comprising applying a therapeutic response to the identified cardiac activity indicative of an abnormal heart rhythm based on the estimated value of the second physiological parameter.

Claims (15)

1. A medical system, the medical system comprising:

a mechanical sensor configured to sense a first physiological parameter signal of a patient; and

processing circuitry configured to:

determining one or more characteristics of the first physiological parameter signal;

applying a machine learning model to the one or more features of the first physiological parameter signal; and

an estimated value of a second physiological parameter is determined based on the application of the machine learning model.

2. The medical system of claim 1, wherein the mechanical sensor is configured to sense a first physiological parameter signal comprising at least one of an endocardial mechanoreceptive signal or an epicardial mechanoreceptive signal encoding cardiac motion data.

3. The medical system of claim 1 or 2, wherein the mechanical sensor comprises an accelerometer.

4. The medical system of any one of claims 1-3, wherein to determine the estimated value of a second physiological parameter, the processing circuit is further configured to generate an estimate of a left ventricular pressure measurement based on the endocardial mechanoreceptive signal or the epicardial mechanoreceptive signal; and/or wherein, to determine the estimated value of the second physiological parameter, the processing circuit is further configured to calculate one or more derivatives of left ventricular pressure measurements based on the endocardial mechanoreceptive signal or the epicardial mechanoreceptive signal.

5. The medical system of any one of claims 1-4, wherein to apply a machine learning model to the one or more features of the first physiological parameter signal, the processing circuit is further configured to apply the machine learning model to accelerometer data and determine an estimate of pressure measurements in a heart chamber based on the application of the machine learning model.

6. The medical system of any one of claims 1-5, wherein to determine the estimated value of the second physiological parameter, the processing circuit is further configured to determine estimated pressure data.

7. The medical system of any one of claims 1-6, wherein the processing circuit is further configured to update the machine learning model based on a comparison of the estimated value to a reference value corresponding to the second physiological parameter; and/or wherein the processing circuitry is further configured to generate the machine learning model by performing a training process on a corpus of first physiological parameter data and second physiological parameter data of a patient population or a subset of patients.

8. The medical system of any one of claims 1-7, wherein the processing circuit is further configured to use the estimated value of the second physiological parameter in an assessment of cardiac synchrony; and/or wherein the processing circuit is further configured to provide therapy to correct cardiac asynchrony using the estimated value of the second physiological parameter.

9. A medical system, the medical system comprising:

communication circuitry coupled to a medical device through a network; and

processing circuitry configured to:

performing a training process on a corpus of cardiac activity data of a patient population or a subset of patients;

Generating a machine learning model based on the training process, wherein the machine learning model is configured to determine an estimate of a second physiological parameter using the first physiological parameter signal; and

the machine learning model is deployed for use in a medical device configured to monitor cardiac activity of a patient.

10. The medical system of claim 9, wherein the processing circuit is further configured to: in response to receiving an estimated left ventricular pressure measurement from the medical device, the machine learning model is updated by comparing the estimated left ventricular pressure measurement for the patient's heart to reference pressure data and adjusting components of the machine learning model based on the comparison.

11. A method, the method comprising:

determining one or more characteristics of a first physiological parameter signal sensed by the mechanical sensor;

applying a machine learning model to the one or more features of the first physiological parameter signal; and

an estimated value of a second physiological parameter is determined based on the application of the machine learning model.

12. The method of claim 11, wherein the mechanical sensor comprises an accelerometer.

13. The method of claim 11 or 12, wherein determining an estimated value of the second physiological parameter further comprises determining estimated pressure data.

14. The method according to any one of claims 11 to 13, further comprising using the estimated value of the second physiological parameter in an assessment of cardiac synchrony; and/or further comprising using the estimate of the second physiological parameter to provide therapy to correct cardiac asynchrony.

15. The method of any one of claims 11 to 14, further comprising identifying cardiac activity indicative of an abnormal heart rhythm based on the estimated value of the second physiological parameter; and/or further comprising applying a therapeutic response to the identified cardiac activity indicative of an abnormal heart rhythm based on the estimated value of the second physiological parameter.