WO2025174909A1 - Concussion subtype identification systems and devices - Google Patents

Abstract

Devices for identifying a concussion subtype of a suspected concussion patient are disclosed. The devices can be used to support classification of concussion patients into subtypes with different prognoses and treatment requirements. Concussion patients can be classified into subtypes based on signals acquired by a device from the patients, such as biosensor signals. The identified concussion subtypes can be associated with etiology, prognosis, and sequelae. Clinicians and concussion patients can benefit from prompt classification, which can support generation of appropriate, personalized treatment and recovery plans, which may in turn support more complete and rapid recovery.

Description

CONCUSSION SUBTYPE IDENTIFICATION SYSTEMS AND DEVICES

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of priority of U.S. Provisional Application No. 63/522,312, filed February 12, 2024, and incorporated herein by reference in its entirety.

TECHNICAL FIELD

[0002] The present disclosure relates to the field of neurological assessment, and specifically, to devices for identifying concussion subtypes in patients.

BACKGROUND

[0003] Concussion is a complex and heterogeneous pathology. But conventional practice treats all patients with concussion similarly. Existing attempts to subtype concussion rely on questionnaires or self-assessments to segment patients into groups, typically based on symptoms reported within 3 to 5 days after injury. Such groupings are not widely accepted or used in standard practice and fail to account for underlying etiology. An inability to group patients based on underlying etiologies can hinder the development of personalized treatment plans for concussion patients.

SUMMARY

[0004] The identification of brain activity -based concussion subtypes can advance the understanding of concussion pathophysiology and improve treatment planning and outcomes. Such identification may preferably be performed proximate to time of injury. Systems consistent with disclosed embodiments can identify concussion subtypes at or after time of injury based on quantitative electroencephalography. The concussion subtypes identified were highly stable, with less than 1% discordance between repeated reclassification runs. Systems consistent with disclosed embodiments exhibited high overall classification accuracy (96.64%) in the discrimination of the subtypes, with sensitivity and precision values per subtype ranging from 95% to 99%. The existence of such physiological subtypes provides evidence for differences in underlying pathophysiology. Observed differences in prognosis and sequalae associated with physiological subtypes provides further evidence for differences in underlying pathophysiology. Measured differences in brain activity can therefore support identification of concussion subtypes indicative of differing underlying pathophysiology. Identification of such concussion subtypes can then support personalized prognosis and treatment optimization.

[0005] The disclosed systems and methods can be used to identifying a concussion subtype in a concussed patient using biosensor signals acquired from the patient. The identified concussion subtype can support a personalized treatment plan. The identified concussion subtype can indicate an etiology of the concussion, which can in turn suggest outcome and potential sequelae, such as mental health conditions (e.g., depression, irritability, anxiety, PTSD, or the like). A clinician can therefore use the identified concussion subtype in planning patient care or monitoring. An identified concussion subtype can also be used for developing or targeting new concussion treatments.

[0006] The disclosed embodiments include a system. The system can include a user device that includes at least one processor and at least one non-transitory computer-readable medium. The at least one non-transitory computer-readable medium can contain instructions that, when executed by the at least one processor, cause the user device to perform operations for identifying a concussion subtype. The operations can include obtaining biosignals acquired from a suspected concussion patient. The biosignals can be associated with brain activity of the suspected concussion patient. The operations can include generating a feature set using the biosignals. The operations can include generating a concussion subtype indication by applying the feature set to a machine learning model trained to identify concussion subtypes using a training dataset. The training dataset used can include observations and corresponding labels, the corresponding labels having been associated with the observations using unsupervised clustering of the observations. The operations can include providing the concussion subtype indication to support diagnosis or treatment planning for the suspected concussion patient.

[0007] The disclosed embodiments include a method. The method can include acquiring biosignals associated with brain activity from a suspected concussion patient. The method can further include generating a feature set using the biosignals. The method can further include associating a concussion subtype with the suspected concussion patient by applying the feature set to a machine learning model trained to identify concussion subtypes using a training dataset. The training dataset used can include observations and corresponding labels, the corresponding labels having been associated with the observations using unsupervised clustering of the observations. The method can include providing medical information associated with the concussion subtype to support diagnosis or treatment planning for the suspected concussion patient.

[0008] The disclosed embodiments include a training system. The training system can include at least one processor and at least one non-transitory computer-readable medium containing instructions. When executed by the at least one processor, the instructions can cause the training system to perform operations. The operations can include obtaining biosignal data. The biosignal data can be acquired from concussion patients and associated with brain activities of the concussion patients. The operations can include generating observations using the biosignal data. Generating a first observation for a first concussion patient can include extracting first features from first biosignal data for the first concussion patient. The operations can include generating a training dataset including the observations and associated class labels. Each class label can correspond to a concussion subtype. The operations can include training a machine learning model to predict a class label from an observation. The operations can include providing the trained machine learning model.

[0009] The foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the claims.

BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying drawings, which are incorporated in and constitute part of this disclosure, together with the description, illustrate and serve to explain the principles of various example embodiments.

[0011] FIG. 1 illustrates a schematic of an exemplary concussion evaluation system 100 for identifying concussion subtypes based on obtained biosignals, consistent with disclosed embodiments.

[0012] FIG. 2A depicts a flowchart of an exemplary method for training a machine learning model to detect concussion subtypes using patient biosensor data, consistent with disclosed embodiments.

[0013] FIG. 2B illustrates a flowchart of an exemplary method for using a machine learning model to detect concussion subtypes using patient biosensor data, consistent with disclosed embodiments.

[0014] FIG. 3A depicts clinical sites at which two prospective longitudinal studies were performed. The data from these studies was used to train a machine learning model to identify concussion subtype.

[0015] FIG. 3B depicts demographic descriptors of the enrolled population of the two studies. [0016] FIG. 4A depicts the final distribution of features and participants for each concussion subtype.

[0017] FIG. 4B depicts the defining quantitative electroencephalography (qEEG) measure sets for each concussion subtype.

[0018] FIG. 5A depicts a t-SNE 2D plot that illustrates how samples from each concussion subtype are distinguished based on the pool of 257 relevant features.

[0019] FIG. 5B and 5C depict accuracy, sensitivity, and precision metrics for the discrimination of concussion subtypes.

[0020] FIGs. 6A to 6E depict polar bar plots of the contributing brain signal features for the identified concussion subtypes, broken out by qEEG feature domains.

[0021] FIG. 7 depicts prevalence estimates for each concussion subtype in the development and test sets.

[0022] FIG. 8A depicts an observed association between subtype and time to return-to-activity event.

[0023] FIG. 8B depicts the cumulative distribution for return-to-activity events for each concussion subtype.

[0024] FIG. 9 depicts an exemplary schematic of a device for acquiring and processing brain electric signals, consistent with disclosed embodiments.

[0025] FIG. 10 depicts an exemplary placement of the device of FIG. 9 on the head of a patient, consistent with disclosed embodiments.

DESCRIPTION OF EMBODIMENTS

[0026] The following detailed description refers to the accompanying drawings. Wherever possible, the same reference numbers are used in the drawings and the following description to refer to the same or similar parts. While several illustrative embodiments are described herein, modifications, adaptations, and other implementations are possible. For example, substitutions, additions, or modifications may be made to the components illustrated in the drawings, and the illustrative methods described herein may be modified by substituting, reordering, removing, or adding steps to the disclosed methods. Accordingly, the following detailed description is not limited to the disclosed embodiments and examples. Instead, the proper scope is defined by the appended claims. Concussion is a complex and heterogeneous disorder. Differing insults can result in differing patterns of structural and functional damage. For example, concussion injuries may have differing pathophysiological profiles reflective of differences in underlying neuronal, axonal, and glial damage and/or microscopic pathology. This heterogeneity of the underlying pathology can be associated with different treatments or prognoses (e.g., different expected post concussive symptoms and long-term sequelae). Differing patterns of structural and functional damage can result in similar clinical symptoms (e.g., headache, nausea, vomiting, balance problems, dizziness, visual problems, fatigue, light sensitivity, noise sensitivity, numbness/tingling, feeling mentally foggy, feeling slowed down, difficult concentrating, difficulty remembering, irritability, sadness, emotional lability, nervousness, drowsiness, changes in amount of sleep, trouble falling asleep, or similar symptoms) but differing patterns of neural activity. Measurements of neural activity can therefore provide better evidence of causal, underlying damage and support more accurate, physiologically based categorization of concussion than clinical features. Categorization of concussion can support improvements in diagnosis and treatment planning. For example, delay in concussion diagnosis and appropriate concussion management have been shown to lead to much slower recovery from concussion. Systems consistent with disclosed embodiments can enable identification of concussion subtype (and therefore provide an indication of underlying pathophysiology) at the time of injury.

[0027] In particular, the temporal resolution of electroencephalography (EEG) can make it sensitive to pathophysiological changes in brain activity resulting from concussion or other traumatic brain injury. Furthermore, features characterizing the EEG signal (i.e. quantitative EEG [qEEG] features) can reflect different mechanisms of underlying brain dysfunction. Such features include those showing shifts in the frequency spectra, disruptions in connectivity between regions and networks and measures of entropy and complexity reflecting neuronal disorganization.

[0028] Consistent with disclosed embodiments, a model can be trained to classify concussion patients into different concussion subtypes. The model can be a machine-learning model. These identified subtypes can be distinguished based on electrophysiological features that capture differing patterns of neural activity. The differing patterns of neural activity in turn reflect the heterogenous underlying pathophysiology of concussion.

[0029] As described herein, different concussion subtypes can be associated with different average recovery times. Some of the identified concussion subtypes exhibited prolonged recovery average time. For example, two identified subtypes (subtype 2, characterized by major disruptions in power features, and subtype 4, characterized by abnormalities in both power and connectivity features) showed longer recovery average periods (three or more weeks) as compared to other subtypes. At the other extreme, another identified subtype (Subtype 3, characterized by disturbances in connectivity measures, coherence, and phase synchrony) showed a rapid recovery trend with almost 50% of participants cleared to return to activity in less than two weeks after injury. [0030] Furthermore, different concussion subtypes appear linked to different underlying pathophysiological mechanisms. These different pathophysiological mechanisms could result in different clinical progression or sequelae, such as prolonged visual system disturbances (e.g., visual convergence disorders) or mental health conditions. For example, one subtype identified herein (Subtype 1) exibited characteristics of disturbances in visual processing (e.g., near point convenrgence). A patient having a concussion of this subtype may require visual therapy. As an additional example, one of the subtypes identified herein (Subtype 3) is characterized by frontal connectivity abnormalities, which have also been observed in non-injured subjects with depression. A patient having a concussion of this subtype may have an elevated risk of developing depression. Likewise, another subtype identified herein (Subtype 4) is charactrized by power abnormalities including those within the theta frequency band (3.5-7.5Hz). Theta activity is prominently generated in the hippocampus, and one well-established finding in PTSD is the presence of hippocampal atrophy. Thus a patient having a concussion of this subtype may have an elevated risk of developing PTSD overtime.

[0031] Systems and methods consistent with the disclosed embodiments can support rapid classification of concussion patients into subtypes with different prognoses and treatment requirements. Clinicians and concussion patients can benefit from such prompt classification, which can support generation of appropriate, personalized treatment and recovery plans, which may in turn support more complete and rapid recovery. For example, based on identified concussion subtype, a clinician may provide guidance to a patient concerning their potential need for visual therapy or elevated risk of PTSD or depression, or more closely monitor the patient for indications of PTSD or depression. As an additional example, based on identified concussion subtype, a clinician may provide an indication of an expected recovery time to the patient. The patient and clinician may then schedule follow-up or monitoring appointments based on the indicated recovery time. Alternatively or additionally, an expected recovery time may provide a benchmark for evaluating the recovery of a particular patient. In at least this manner, the disclosed systems and methods can improve the treatment of concussion patients.

[0032] FIG. 1 illustrates a schematic of an exemplary concussion evaluation system 100 for identifying concussion subtypes based on obtained biosignals, consistent with disclosed embodiments. The concussion evaluation system 100 can include user device 102, and measurement device(s) 112. The user device 102 may be communicatively (directly or indirectly) coupled to measurement device(s) 112. Consistent with disclosed embodiments, measurement device(s) 112 can be configured to acquire and provide biosignals to user device 102. In some embodiments, concussion evaluation system 100 can be portable. As may be appreciated, the disclosed embodiments are not limited to a concussion evaluation system 100 that includes a single user device 102 and a measurement device. The concussion evaluation system 100 may include any number of user devices and any number of measurement devices. In some embodiments, concussion evaluation system 100 can be configured to receive biosignals from another system, which may include measurement device(s) 112.

[0033] Consistent with disclosed embodiments, user device 102 may be, for example, a computing system, such as a handheld device (e.g., a special purpose medical device, such as a portable EEG machine, or smartphone), a wearable device, tablet, laptop, desktop, workstation, computing cluster, or cloud computing platform. In some embodiments, user device 102 can be configured using an application (e.g., a downloadable “app”) to identify concussion categories. In some embodiments, user device 102 can be configured to use line power and/or an internal power source, such as a battery or fuel cell. In some embodiments, user device 102 can be configured to power measurement device(s) 112. In some embodiments, the user device 102 may be configured to process biosignals received from measurement device(s) 112 (e.g., into EEG data suitable for subsequent analysis, or the like). In some embodiments, such processing can include one or more of signal conditioning, filtering, artifact rejection, feature extraction, or the like.

[0034] Consistent with disclosed embodiments, user device 102 may include processor(s) 104, memory 106, display 109, and network interface 110. Processor(s) 104 can include suitable logic, circuitry, and/or code that enable processing data and/or controlling operations of the user device 102. In this regard, processor(s) 104 can be enabled to provide control signals to various other components of the user device 102. Processor(s) 104 can also control transfers of data between various portions of the user device 102, or between user device 102 and other devices or systems. Processor(s) 104 can be configured to implement an operating system or may otherwise execute code to manage operations of the user device 102.

[0035] Consistent with disclosed embodiments, memory 106 can include suitable logic, circuitry, and/or code that enable storage of various types of information such as received data, generated data, code, executable instructions, and/or configuration information, consistent with disclosed embodiments. The memory 106 may include, for example, random access memory (RAM), read-only memory (ROM), flash, and/or magnetic storage.

[0036] Consistent with disclosed embodiments, display 109 can provide an input interface (e.g., touch input or stylus) and/or an output interface (e.g., visual output) between user device 102 and a user, consistent with disclosed embodiments. Display 109 may be configured to display a visual output to the user (e g., graphics, text, icons, video, or the like). Display 109 may be at least one of an LCD (liquid crystal display), an LPD (light emitting polymer display), or an LED (light emitting diode), organic LED (OLED) type displays, although other display types are considered.

[0037] Consistent with disclosed embodiments, network interface 110 can include suitable logic, circuitry, and/or code that enables wired or wireless communication, such as between the user device 102 and other device(s), consistent with disclosed embodiments. The network interface 110 may include, for example, one or more of a BLUETOOTH communication interface, an NFC interface, a ZIGBEE communication interface, a WLAN communication interface, a LAN interface, a USB communication interface, or generally any communication interface.

[0038] Consistent with disclosed embodiments, measurement device(s) 112 can be configured to acquire data concerning a patient. The acquired data can be or include biosignals, consistent with disclosed embodiments. Such biosignals can be or include electrophysiological data (e.g., electroencephalography (EEG) or qEEG data, electrocardiogram (ECG) data, electrocorticography (ECoG), evoked potential data, or the like), imaging data (e.g., nuclear magnetic resonance spectroscopy (NMR) data, magnetoencephalography (MEG) data, positron emission tomography (PET) data, functional magnetic resonance imaging (fMRI) data, or the like) or other suitable biosignals. For example, measurement devices consist with disclosed embodiments can include (but are not limited to) electrode arrays; EEG, ECG, or ECoG devices; heart rate monitors; accelerometers, pulse oximeters, digital imaging devices, NMR, MEG, MRI, PET devices, or the like. For example, measurement device(s) 112 may be or include an array of electrodes. These electrodes can be configured and adapted to measure electrical activity in the body of a patient, when suitably disposed on the patient. In some embodiments, the electrodes can be electroencephalogram (EEG) electrodes configurable to measure brain activity. [0039] In some embodiments, the acquired data can further include patient physiological data (e.g., heart rate, breath rate, tidal volume, blood pressure, neuro-ophthalmic data, or the like, or statistics or variability or the same); patient motion data; environmental condition data (e.g. sound data, light data, weather data, or the like); or the like. For example, such a measurement device may include one or more of: an accelerometer for detecting user acceleration, an audio sensor (e.g., microphone) for detecting sound, an optical sensor for detecting light, a camera for detecting patient eye movements, and/or other suitable sensor(s) configured to output signals indicating patient physiological state, patient motion, and/or environmental conditions.

[0040] In some embodiments, one of measurement device(s) 112 can be configured to provide a stimulus. The stimulus can be selected to evoke a response in the patient (e.g., an evoked potential test or the like). The response can be measured by the measurement device providing the stimulus, or another measurement device. For example, such a measurement device may include one or more of: an audio emitter (e.g., a speaker, earbud, headphone, or the like) for providing an auditory stimulus, a stimulation electrode for providing an electrical stimulus, a mechanical stimulation device for providing haptic stimulus, a thermal stimulation device for providing thermal stimulus, a visual stimulation device (e.g., a light source, displays, or the like) for providing visual stimulus, or the like.

[0041] In some embodiments, measurement device(s) 112 can be self-powered. For example, measurement device(s) 112 can be configured to use line power and/or an internal power source, such as a battery or fuel cell. In some embodiments, measurement device(s) 112 can be powered by user device 102 (or another system). In some embodiments, measurement device(s) 112 can process the acquired data provided to user device 102. Such processing can include one or more of signal conditioning, filtering, artifact rejection, feature extraction, or the like. [0042] As may be appreciated, the disclosed embodiments are not limited to the concussion evaluation system depicted in FIG. 1. A concussion evaluation system consistent with disclosed embodiments can include additional or fewer components than concussion evaluation system 100. For example, such a system may include multiple measurement device(s) 112 or may not include measurement device(s) 112 (e.g., instead obtaining the acquired data from a database or another system). In some embodiments, one or more of the included measurement device(s) 112 can be integrated into user device 102. For example, display 109 can be used to provide visual stimuli and user device 102 can include a camera for measuring patient eye movements. As an additional example, earbuds communicatively connected to user device 102 can provide auditory stimuli and a separate EEG device can detect evoked auditory potentials and provide auditory potential data to user device 102. As a further example, a concussion evaluation system may not include display 109 or network interface 110. For example, user device 102 may be configured to communicatively connect to another system (e.g., using a wired or wireless connection) and provide data to that system for display or transmission over a network. In some embodiments, one or more measurement device(s) 112 can be separate from user device 102.

[0043] FIG. 2A depicts a flowchart of an exemplary method 210 for training a machine learning model to detect concussion subtypes using acquired patient data, consistent with disclosed embodiments. Method 210 can be performed by a computing system, such as a laptop, desktop, workstation, computing cluster, or cloud-computing platform. Method 210 can include obtaining training data, generating a training dataset using the training data, and training a machine learning model using the training dataset. While described principally with respect to biosensor data (and, in particular, EEG data), the disclosed embodiments are not so limited. Other acquired data can be additionally or alternatively used, consistent with disclosed embodiments. For example, patient physiological data or motion data (or features based thereon) can be included in the training dataset.

[0044] In step 211 of method 210, the computing system can obtain training biosensor data, consistent with disclosed embodiments. In some embodiments, the computing system can obtain database(s) or file(s) containing at least some of the training biosensor data. In some embodiments, the computing system can obtain at least some of the training biosensor data from a measurement device (e.g., one of measurement device(s) 112). The measurement device can be configured to acquire the biosensor data from patients. For example, the biosensor data can be EEG data and the measurement device can be an EEG machine. The EEG machine can be configured to receive EEG signals from EEG electrodes disposed on the head of the patient. The training biosensor data can include samples associated with individual patients. For example, recording(s) of EEG data from a patient can constitute a sample associated with that patient. [0045] In some embodiments, obtaining the training biosensor data can include preprocessing steps such as removing artifacts and noise from the biosensor data. For example, physiological and non-physiological contamination (e.g., eye movement, electromyography muscle activity, or the like) can be removed (e.g., by inspection, filtering, real-time artifact detection algorithms, or another suitable method). Such preprocessing steps can include converting the samples to a standard format. For example, a sample can be converted to a standard length, sample rate (e.g., using upsampling or downsampling), standard bandwidth or frequency range (e.g., using filtering, such as bandpass filtering) or the like.

[0046] In step 212 of method 210, the computing system can generate observations by extracting features from the training biosensor data, consistent with disclosed embodiments. The extracted features can constitute observations. For example, the computing system can extract a set of features from a training biosensor data observation associated with a patient (e.g., a recording of EEG data for the patient). The extracted set of features can constitute an observation associated with the patient.

The features can be selected to quantify brain activity in different brain regions and/or over different time scales or frequencies. In some embodiments, the features can be qEEG features. Such qEEG features can be grouped into measure sets including at least one of signal power (e.g., absolute, relative, and the like), mean frequency, connectivity (e.g. asymmetry, coherence, phase lag, phase synchrony, or the like), complexity (e.g., fractal dimension and scale-free activity, or the like), information theory (e.g., entropy or the like), or the like. In some embodiments, the extracted features can be standardized over patients. For example, the extracted features can be z-transformed over patient demographic characteristics or combinations thereof (e.g., age, sex, or other relevant characteristics). In some embodiments, features can be extracted as disclosed in “Classification of traumatic brain injury severity using informed data reduction in a series of binary classification algorithms,” by Prichep LS, Jacquin AE, Filipenko I, et al., IEEE Trans Neural Syst Rehab Eng. 2012;20(6): 806-822, or “A multimodal biomarker for concussion identification, prognosis and management” by Jacquin AE, Kanakia S, Oberly D, Prichep LP., Comput Biol Med. 2018;102:95-103, and incorporated herein by reference.

Examples of suitable coherence, phase, and amplitude difference are given in “An EEG severity index of traumatic brain injury” by Thatcher RW, North DM, Curtin RT et al. J Neuropsychiatry Clin. Neurosci. 2001 ; 13(l):77-87 and incorporated herein by reference. Examples of suitable connectivity measures are given in “Changes in Functional Brain Networks following Sports- Related Concussion in Adolescents” by Virji-Baul N, Hilderman CGE, Makan N, et al. J Neurotrauma. 2014;31(23): 1914-1919 and incorporated herein by reference. Examples of additional suitable measures are given in “Diffusion Tensor Imaging Indicators of White Matter Injury Are Correlated with a Multimodal Electroencephalography-Based Biomarker in Slow Recovering, Concussed Collegiate Athletes” by Wilde E, Goodrich-Hunsaker NJ, Ware AL, et al. J Neurotrauma. 2014;37(19):2093 -2101 and incorporated herein by reference. Examples of suitable information theory measures are given in “Differential Effect of First versus Second Concussive Episodes on Wavelet Information Quality of EEG” by Slobounov S, Cao C, and Sebastianelli W. Clin. Neurophysiol. 2009; 120(5):862-867 and incorporated herein by reference. Additional examples of suitable features are given in US Patent Publication No. 2011/0112426 and incorporated herein by reference. In some embodiments, the number of extracted features can be reduced to make the feature set more computationally tractable and/or reduce likelihood of over-training of the machine learning model. In some embodiments, features that were not replicable in normally functional (non-concussed) individuals were discarded. In some embodiments, such extraction can include retaining features that satisfy a variation criterion for the patient population (e.g., as such variation can support discrimination between concussion subtypes). Satisfaction of the variation criterion can be determined using resampling techniques. For example, a randomized permutation test can be used to compute significance values for features. Based on the significance values, the features can be retained or discarded.

[0047] In step 213 of method 210, the computing system can associate observations with class labels, consistent with disclosed embodiments. In some embodiments, observations can be clustered to identify similar groups of observations. The disclosed embodiments are not limited to any particular clustering method. As may be appreciated, depending on the clustering technique, the number of clusters can be predetermined or can be determined based on the results of the clustering (or using other algorithms adapted to optimize a number of clusters according to some optimization criterion). Each cluster can then be associated with a class label. In some embodiments, a class label can indicate a concussion subtype, etiology or type of injury, prognosis, potential sequelae, or the like. A class label can be assigned to a cluster based on an assessment of the observations contained in the cluster. In some embodiments, class labels can lack indicative value beyond distinguishing between clusters.

[0048] In some embodiments, spectral co-clustering can be used to associate class labels with observations. Spectral co-clustering can be performed on a matrix having the observations as a first dimension and the extracted features as a second dimension. The spectral co-clustering can partition the rows and columns of a matrix, finding subsets of rows that change similarly over a subset of columns, or subsets of rows that have similar values across a subset of columns. The resulting partitions of rows and columns are referred to as biclusters. Spectral co-clustering can assume that the data has a checkerboard-like structure with blocks of high-expression levels and low-expression levels. Spectral co-clustering can find these distinctive checkerboard patterns using eigenvectors and singular value decomposition (SVD) of matrices. The singular vectors for rows and columns can then be clustered using K-means. Every row and column can be included only in one bicluster, with the resulting structure being block-diagonal. Each bicluster can be associated with a class label.

[0049] As described herein, in some embodiments, class labels may be associated with distinct or related concussion etiologies. However, given the unsupervised nature of the clustering, in some embodiments such an association cannot be guaranteed. In such embodiments, one or more class labels can correspond to “all-other”, “multi-causal”, “not classified”, “non-specific”, or the like. [0050] Consistent with disclosed embodiments, medical information (e.g., etiologies; clinical symptoms; recovery time, outcome, or trajectory; sequela; or the like) can be associated with class labels based on the observations contained in each class. For example, recovery time statistics can be determined for a class based on the patients corresponding to the observations in the class. Likewise, visual system disturbances, such as disturbances in visual near point convergence, can be determined for a class based on the patients corresponding to the observations in the class. Likewise, elevated risk of mental health conditions can be determined for patients in a class. Similarly, imaging data, biomarker data, pathology or histology data can be analyzed by class. In this manner, the classes identified in step 213 using unsupervised learning can be mapped to underlying pathophysiology and clinical consequences thereof.

[0051] In step 214 of method 210, the computing system can generate a final feature set for use in training a machine learning model, consistent with disclosed embodiments. In some embodiments, the final feature set can be generated by eliminating from an initial feature set (e.g., the output of step 212) less-predictive features or combinations of features. In some embodiments, the final feature set can be generated by including in the final feature set more- predictive features or combinations of features in the initial feature set. Such feature selection can be performed manually (e.g., by a data scientist reviewing values or statistics of the features), semi-automatically (e.g., through application of a feature selection method and manual review of the features selected), or automatically using known statistical methods. The disclosed embodiments are not limited to any particular method of feature reduction. In some embodiments, the computing system can perform recursive feature elimination, as described in “Gene selection for cancer classification using support vector machines,” by Guyon I, Weston J, Barnhill S, Vapnik V, Machine learning. 2002 Jan;46:389-422, and herein incorporated by reference. Such recursive feature elimination can include steps of training a classifier using an input feature set (e.g., the initial feature set or subset thereof), computing a ranking criterion for each feature used by the classifier (e.g., a change in a performance measure of the classifier when the feature is disregarded), and removing feature(s) from the input feature set based on the ranking criterion (e.g., the lowest-ranked feature can be removed). These steps can be repeated until a termination condition is satisfied (e.g., a feature set of a predetermined size is obtained, a performance condition of the resulting classifier is satisfied, or another suitable condition). The disclosed embodiments are not limited to using a particular classifier for performing recursive feature elimination. In some embodiments, the classifier can be a support vector machine, a linear or logistic regression classifier, a decision tree or random forest, or the like. In some embodiments, the classifier can be an ensemble of other, less-predictive classifiers.

[0052] In some embodiments, recursive feature elimination can be performed multiple times using different subsets of the training data or different random seeds. As a result, different feature sets can be generated. In some embodiments, the different feature sets can be combined to form the final feature set. For example, a feature can be included in the final feature set based on whether all the different feature sets included the feature, the number of different feature sets that included the feature, the importance of the feature in the different feature sets, or another suitable criterion. In some embodiments, the feature set with the highest value of a performance measure can be retained (e.g., the feature set yielding a classifier with the highest classification accuracy, precision, recall, Fl-score, or the like).

[0053] In step 215 of method 210, the computing system can train a machine learning model to predict a class label for an observation, consistent with disclosed embodiments. The machine learning model can accept values of the final feature set and can output an indication of the class label. The disclosed embodiments are not limited to using a particular machine learning model. In some embodiments, the machine learning model can be a support vector machine(e.g., a one- vs-one (classes) or one-vs-all classes support vector machine) , a linear or logistic regression classifier (e.g., a multinomial logistic regression classifier), a decision tree or random forest, or the like. In some embodiments, the machine learning model can be an ensemble of other, less- predictive classifiers. In some embodiments, the machine learning model can be a logistic regression classifier using regularization (e.g., L2-norm regularization) and/or balanced class learning (e.g., using weights adjusted inversely proportional to class frequencies in the data). As may be appreciated, the performance of the machine learning model can be evaluated using holdout observations, or using another suitable method.

[0054] In step 215 of method 210, the computing system can provide the machine learning model, consistent with disclosed embodiments. Providing the machine learning model can include storing the machine learning mode in a memory accessible to the computing system, providing the machine learning model to another computing system, configuring or provisioning a user device (e.g., user device 102 or the like) with the machine learning model, making the machine learning model available for queries from other systems, or the like.

[0055] For example, the computing system (or another system that receives the machine learning model the computing system) can configure or provision a portable EEG device with the machine learning model. The computing system (or other system) can also configure the portable EEG device to generate the final feature set from acquired biosignal data (e.g., performing preprocessing and feature extraction as necessary). So configured or provisioned, the portable EEG device can be capable of classifying concussion subtypes in suspected concussion patients using biosignals acquired from the suspected concussion patents. Such a portable EEG device can be used in clinical or non-clinical settings (e.g., schools or athletic facilities, workplaces, publica areas, military base or field use, or other suitable settings).

[0056] As an additional example, the computing system (or another system that receives the machine learning model the computing system) can respond to queries received (e.g., using an application programming interface, remote procedure call, web service, or the like) from other systems (e.g., systems associated with clinical or non-clinical settings, as described herein). The queries can include biosensor data, or final feature sets, or the like. The computing system (or other system) can use the queries and the machine learning model to generate concussion subtype indications, which may be provided in response to the queries.

[0057] In some embodiments, the classifier of step 214 and the machine learning model of step 215 can be of different types or different architectures. For example, the classifier of step 214 can be simpler (e.g., including fewer layers, weights, or values; requiring less computation or memory for training or inference; or the like) or generate a prediction faster than the machine learning model of step 215. In some embodiments, the classifier of step 214 and the machine learning model of step 215 can be of the same type.

[0058] The disclosed embodiments are not limited to embodiments that perform every step of method 210. In some embodiments, steps can be removed or combined, or new steps can be added. For example, steps 214 and 215 can be combined. One of the classifiers used to generate the final feature set (e.g., the classifier with the greatest accuracy among feature sets having less than a threshold number of features) can be the machine learning model of step 215. As an additional example, step 214 can be omitted, and the machine learning model can be trained using the feature set generated in step 212. As a further example, steps 211 and 212 may be omitted and the machine learning model can be generated using observations provided by another system.

[0059] FIG. 2B illustrates a flowchart of an exemplary method 220 for using a machine learning model to detect concussion subtypes using patient biosensor data, consistent with disclosed embodiments. Method 220 can be performed by a handheld device as described herein with regards to FIGs. 9 and 10, or another portable device (e.g., another special purpose medical device, such as a portable EEG machine, or smartphone), or another suitable computing system (e.g., a wearable device, tablet, laptop, desktop, workstation, computing cluster, or cloud computing platform). For example, method 220 can be performed by a user device (e.g., user device 102, or the like). Method 220 can include obtaining patient data, generating a patient observation using the patient data, and applying the patient observation to a machine learning model trained to output indications of concussion subtypes. Consistent with disclosed embodiments, the machine learning model can be generated using method 210, or another method. Consistent with disclosed embodiments, the computing system can generate the machine learning model, or obtain the machine learning model from a user or another computing system. While described principally with respect to a patient observation including biosensor data (and, in particular, EEG data), the disclosed embodiments are not so limited. Other acquired data can be additionally or alternatively used, consistent with disclosed embodiments. For example, patient physiological data or motion data (or features based thereon) can be included in the patient observation.

[0060] In step 221 of method 220, the computing system can obtain patient biosensor data, consistent with disclosed embodiments. In some embodiments, the computing system can obtain database(s) or file(s) containing the patient biosensor data. In some embodiments, the computing system can obtain at least some of the patient biosensor data from a measurement device. The measurement device can be configured to acquire the biosensor data from the patient. For example, the biosensor data can be EEG data and the measurement device can be an EEG machine. The EEG machine can be configured to receive EEG signals from EEG electrodes disposed on the head of the patient. The patient can be a patient suspected of having a concussion.

[0061] In some embodiments, obtaining the patient biosensor data can include preprocessing steps such as removing artifacts and noise from the biosensor data. For example, physiological and non-physiological contamination (e.g., eye movement, electromyography muscle activity, or the like) can be removed (e.g., by inspection, filtering, real-time artifact detection algorithms, or another suitable method). Such preprocessing steps can include converting the patient biosensor data to a standard format. For example, the patient biosensor data can be converted to a standard length, sample rate (e.g., using upsampling or downsampling), standard bandwidth or frequency range (e.g., using filtering, such as bandpass filtering) or the like.

[0062] In some embodiments, the preprocessing steps performed can be the same as the preprocessing steps used to generate the training data for the machine learning model (e.g., the preprocessing steps performed in step 211).

[0063] In step 222 of method 220, the computing system can generate a patient observation by extracting features from the patient biosensor data, consistent with disclosed embodiments. The extracted features can constitute observations. For example, the computing system can extract a set of features from a training biosensor data observation associated with a patient (e.g., a recording of EEG data for the patient). The extracted set of features can constitute an observation associated with the patient. [0064] In some embodiments, the extracted features can be the same as the extracted features included in the training data for the machine learning model (e.g., the features in the final feature set used in step 215).

[0065] In step 223 of method 220, the computing system can apply the patient observation generated in step 222 to the trained machine learning model, consistent with disclosed embodiments. The trained machine learning model can output a classification for the patient observations. The output classification can indicate a concussion subtype for the patient.

[0066] In step 224 of method 220, the computing system can provide the output classification, consistent with disclosed embodiments. The output classification can indicate a one of the concussion subtypes for the patient (or, for example, multiple concussion subtypes and assorted rankings or likelihood scores). Additional information can be provided together with the output classification. Such additional information can include extracted feature(s) and value(s) thereof, information about the extracted feature(s), biosensor information or values, treatments (or treatment schedules) associated with an indicated concussion class for the patient. In some implementations, the indication may include, based on the output classification, an estimated time-period or time window until the patient can resume various activities, such as sports. In some embodiments, the output classification (and any additional information) can support diagnosis of a concussion subtype in the patient.

[0067] The disclosed embodiments are not limited to a particular recipient of the output classification. In some embodiments, the computing system can provide the output classification to the patient; a relative, guardian, or caregiver of the patient; a clinician; or another person involved in the medical care or treatment of the patient. [0068] The disclosed embodiments are not limited to a particular method of providing the output classification. In some embodiments, the computing system can provide the output classification using a display, printer, audio output or the like of the computing system. In some embodiments, the computing system can provide the output classification to another system (e.g., a system of the patient, clinician, or the like). The other system can then provide the output classification using a display, printer, audio output or the like.

[0069] The disclosed embodiments are not limited to a particular format for providing the output classification. In various embodiments, the output classification can be or include a pop-up message, a webpage, an email, a message, an image, or video. In some embodiments, providing the output classification can include storing the output classification in a file or database. For example, the computing system can update a medical record of the patient using the output classification (or provide instructions to another system to update a medical record of the patient using the output classification).

[0070] The disclosed embodiments are not limited to embodiments that perform every step of method 220. In some embodiments, steps can be removed or combined, or new steps can be added. For example, steps 221 and 222 may be omitted and the output classification can be generated using a patient observation provided by another system.

[0071] GENERATING A MACHINE LEARNING MODEL

[0072] Two prospective longitudinal studies were performed at the clinical sites depicted in FIG. 3A. Study participants were ages 13 to 70 years and clinically deemed concussed at their respective sites. Participants were assessed with a handheld EEG device within 120 hours of injury. Retum-To- Activity (RTA) was clinically determined by site standard practice (gradual return protocols). All participants signed written informed consent and for minors, parental written informed consent and adolescent assent were also obtained. Race and ethnicity were self- identified based on single-choice entries. The studies followed STARD reporting guidelines and were registered on clinicaltrials.gov. This retrospective cohort study followed STROBE guidelines for observational studies.

[0073] Inclusion criteria

[0074] Participants with concussion consisted of males and females between the ages of 13 and 70 years, who met the study definition of concussion and had a Glasgow Coma Scale (GCS) score >14 at the time of injury and no hospital admission due to either head injury or collateral injuries for >24 hours.

[0075] Exclusion Criteria

[0076] Participants with forehead, scalp or skull abnormalities or whose clinical condition would not allow electrode placement; current psychoactive prescription medications taken daily (with the exception of medications for attention deficit disorder); history of brain surgery or neurological disease (including multiple sclerosis, Alzheimer’s, or Parkinson’s disease); pregnant women; acute intoxication; active fever greater than 100°F (37.8°C); inability to speak or read English. Subjects with a concussion were excluded if they lost consciousness for >20 minutes, or if there was evidence of abnormality visible on head CT related to the injury.

[0077] Study definition of Concussion and Return to Activity

[0078] Participants with concussion were defined as those subjects who had a witnessed head impact and who, by site guidelines, were restrained from normal activity for five or more days. RTA determination (number of days to cleared to resume activity date) was made in accordance with a gradual/graduated RTA protocol across multiple days, at the end of which a subject was cleared to return to activity/play. For the non-sport concussion subjects (13.2 %), the RTA was defined by physician standard of care. For college and high school -based sites, gradual RTA protocols conformed to National Collegiate Athletic Association and policy guidelines.

[0079] Clinical Assessments

[0080] Study participants were evaluated with three sections of the Sports Concussion Assessment Tool - 3^rd Edition (SCAT-3) or 5^th Edition (SCAT-5): 1) GCS; 2) 22 item Concussion Symptom Inventory (CSI) self-rated on a Likert scale (0-6 per item, total score range 0-132); and 3) Standard Assessment of Concussion (SAC): a brief neurocognitive screening tool (total score range 0-30). History of head injury and concussion was also acquired. [ 0081 ] EEG Data Acquisition

[0082] Ten minutes of eyes closed resting EEG data was collected. The EEG data were recorded using a disposable headset which included Fpl, Fp2, F7, F8, AFz, Al, and A2 locations of the expanded International 10-20 Electrode Placement System, re-referenced to linked ears, and all electrode impedances were below 10 kQ throughout the recording. Data were acquired at a sampling rate of 1 kHz. Amplifiers had a band pass filter from 0.3 to 250 Hz (3 dB points) and down sampled to 100 Hz for feature extraction.

[0083 ] EEG Data Processing and Quantitative EEG Feature Sets

[0084] Physiological and non-physiological contamination (e.g., eye movement, electromyography muscle activity) was removed from the 10-minute EEG recording using realtime artifact detection algorithms, resulting in 1-2 minutes of artifact-free data for each participant. A set of >6,000 qEEG features was extracted afterward and z-transformed with respect to age expected normal values. The extracted features quantify characteristics of the electrical brain activity of different regions and frequency bands (1.5 to 45 Hz), expressed through measure sets as described herein. [0085] A model including too many features may be susceptible to overfitting. Such a model may perform well on a training dataset, but may exhibit inferior performance on real-world or out-of-training data. Furthermore, the usability and interpretability of a model can decrease as the number of features increases. Accordingly, a two-stage feature reduction was conducted to reduce the number of features. First, only features demonstrated to be replicable over the time period of the study were included for the next stage of feature reduction. Secondly, features showing variable ranges within the injured population were included, as such heterogeneity would contribute to the separation of subtypes of concussion. Heterogeneity was computed based on one hundred bootstrapped repetitions of a randomized permutation test using p-values from the Kolmogorov- Smirnov statistic. Features deemed non-significant in more than 5% of the repetitions were discarded. As a result of the feature reduction, 471 qEEG features were retained. [0086] Machine Learning Model Generation / Training

[0087] The training dataset included 771 injured subjects. An initial random division split the cohort into 600 subjects for training (-85%) and 171 subjects for testing.

[0088] As may be appreciated, generating the training dataset included an unsupervised step of identifying physiologically based concussion subtypes. A supervised classification process was then used to label the training examples using the identified subtypes.

[0089] Spectral co-clustering was used on the training split to detect distinct partitions (subtypes) of EEG activity. Spectral co-clustering enabled unsupervised identification of classes in the measured EEG activity. This method found subsets of rows that change similarly over a subset of columns and/or subsets of rows that have similar values across a subset of columns. The resulting partitions of rows and columns are referred to as biclusters. The method generated a matrix having a checkerboard -like structure, with blocks of high-expression levels and low- expression levels. These blocks were identified using eigenvectors and singular value decomposition of matrices. The singular vectors for rows and columns were then clustered using K-means. Each row and column were included only in 1 bicluster, with the resulting structure being block-diagonal. In this manner, the spectral co-clustering simultaneously partitioned the rows and columns of a matrix of the training split into biclusters. To verify the stability of the biclusters, 100 multi start runs of the co-clustering algorithm were performed using random seeds to initialize partitions.

[0090] After reaching stable partitions in the training set, supervised classification was performed using the bicluster memberships as class labels. To further increase usability and interpretability, and reduce the risk of overfitting, an informed data reduction was performed using a stepwise feature elimination method that finds the optimal set of features for a given classification function. In each iteration, the classification function is trained using the remaining features for the training data. The importance of the features for the resulting classifier is determined, and the least-important feature removed from the set of remaining features. This process can be repeated until a stopping criterion (e.g., based on the number of features, the performance of the classifier, or the like) is satisfied. The method started with the full set of 471 features and iteratively removed the least relevant feature. The feature elimination method was run on a 100 multistart setup. Each run performed a random five times 5-fold stratified cross- validation and two functions were used, support vector machines and logistic regression, to guide the search process.

[0091] A final classification algorithm was derived over training data using the feature subset with highest classification accuracy. This classifier was derived from a logistic regression using an L2-norm regularization and balanced class learning through adjusting weights inversely proportional to class frequencies in the data. The hold-out subjects were classified into subtypes by this algorithm. Any statistical significance reported in the results section was fixed to a p- value of 0.05 unless explicitly stated otherwise.

[0092] Results

[0093] FIG. 3B depicts demographic descriptors of the enrolled population of the two studies. Overall, there were 771 concussed participants, 56% male, with a mean age of 20.16 (5.75) years, a mean GCS of 15 (0.06), and median time since injury of 50.03 hours. Return to activity was categorized as rapid recovery (<14 days, 47.73%) or prolonged recovery (> 14 days, 36.96%). There were 15.30% unexpected failures to follow-up for RTA, approximately half of which were due to the COVID- 19 pandemic (55 out of 118) reducing the ability for participants to return for follow-up evaluations. Race and ethnicity distributions followed same percentages as those published by the 2020 US Census Report.

[0094] Subtypes (data profiles or phenotypes) found by the co-clustering runs over the training set were aggregated to analyze stability and identify possible noisy features or subjects. Five stable distinctive data profiles of subjects and features were identified across all the runs, with an average pair-wise consensus score of 98.87% ± 1.19.

[0095] FIG. 4A depicts the final distribution of features and participants for each concussion subtype. Only five subjects (out of the original 600 in training, < 1%) and two features (out of 471, <1%) were systematically assigned into different subtypes over all runs. These subjects and features were deemed “noisy” and removed from further analyses, resulting in 595 participants and 469 features being used to describe the subtypes.

[0096] FIG. 4B depicts the defining qEEG measure sets (those that contributed most to the profile) for each subtype. The distribution of the statistically significant features is segmented by qEEG measure set (e.g., power, power ratios, connectivity, and complexity). The y-axis shows the distribution within the subtype by measure set (total to 100). As can be observed, distinct patterns of qEEG measure sets characterize the different subtypes, supporting a conclusion that the different subtypes correspond to different underlying pathophysiologies.

[0097] As described herein, each observation was labeled using the associated subtype membership from the co-clustering analysis. A subset of 257 features were identified out of the input pool of 469 features (a 54.8% reduction).

[0098] FIG. 5A depicts a t-SNE 2D plot that illustrates how samples from each concussion subtype are distinguished based on the pool of 257 relevant features. A t-SNE 2D plot is a visualization technique that reduces the dimensionality of a set of points from the original descriptors to just two axes by keeping the relative distances from the original data space. It can be seen in FIG. 5A how each subtype occupies a different part of the plotting space forming separable groups. Quantitatively, support vector machines and logistic regression were tested as classification models on a five-fold cross-validation scheme. Logistic regression showed the best overall performance with a 96.64% classification accuracy. FIG. 5B depicts a confusion matrix of a five-fold cross validation estimation for a logistic regression classifier evaluated on the training data set (e.g., 257 features, 595 subjects labeled with the co-clustering outputs). FIG. 5C depicts sensitivity (e.g., true positive rate) and precision (positive predictive value) for each subtype. The value of these figures of merit ranged between 95% and 99%.

[0099] FIGs. 6A to 6E depict polar bar plots of the contributing brain signal features for the identified concussion subtypes, broken out by qEEG feature domains (e.g., measure set). In such polar bar plots, the distance from the origin represents the magnitude of the abnormality in both the positive direction (excess relative to age expected normal values) and the negative direction (deficit relative to age expected normal values). Here, the polar bar charts display the average z- score for a common set of relevant qEEG features contributing to the characterization of the subtypes, segmented by measure set (e.g., absolute power, mean frequency, asymmetry, complexity (scale free, fractal dimension, and entropy), coherence, phase stability, phase synchrony, power ratio, and relative power). This common set of features does not represent all features used in the co-clustering, nor necessarily those which were most significant in differentiating between subtypes. However, when comparing the polar bar charts for all five reported subtypes, distinctive overall patterns of abnormalities characterizing each subtype are clearly observed.

[0100] For example, Subtypes 1 (FIG. 6A) and 4 (FIG. 6D) show opposite values with respect to relative and absolute power, power ratios, mean frequency, and complexity. Thus, while Subtype 1 is characterized by extreme excesses for complexity features and mean frequency, and deficits of absolute and relative power, Subtype 4 shows the opposite pattern. As an additional example, Subtypes 2 (FIG. 6B) and 3 (FIG. 6C), other domains showed opposite activities (e.g., phase synchrony, phase stability, and coherence). In both comparisons, the differential activity is highly significant with absolute differences over two z-score units (significance level of a group average z-score can be estimated by considering the square root of the size of the group, a z- score difference of 0.5 is associated with /?<0.0005). The remaining Subtype 5 polar bar chart showed no differential brain activity compared to the other four groups with mean z-values close to 0 suggesting an unspecific form of injury (see FIG. 6E).

[0101] Prevalence estimations between development and test data demonstrated similar rates. It was noted that Subtype 4 was the most common concussion subtype in both development and test populations (22.8% and 22.9%, respectively), and that Subtype 1 was one of the least common subtypes (15.13% and 16.37%, respectively). FIG. 7 depicts prevalence estimates for each concussion subtype in the development and test sets. Rank values display ordinal estimations of occurrence based on prevalence (from development and validation, and final average positions).

[0102] Subtype membership and clinical symptoms

[0103] A clear relationship was observed between subtype membership and outcome (days to clinical recovery, RTA). FIG. 8A depicts an observed association between subtype and time to return-to-activity event. The counts and percentages are determined over subtype valid data. Key values are highlighted in bold for reference. Asterisks under each subtype refer to statistically significance differences (** p-value <0.01, * p-value < 0.05). RTA days were aggregated by weeks (one, two, three, four or more) to analyze group patterns and minimize site deviations. From the 766 subjects in training and test splits, RTA values for 652 participants were available, with an outlier upper bound of three standard deviations over the mean of 85.94 days. Outliers beyond 86 days were removed, reducing the number of valid values to 640 with a median of 16 days (11 days for the 25^th percentile, 23 days for the 75^th percentile). It can be observed that Subtype 2 and Subtype 4 showed prolonged recovery with RTA values over four weeks in 40% to 35% of the cases, respectively. At the other extreme, Subtype 3 and Subtype 5 presented quick recovery patterns on average with an aggregate 40% or more subjects being cleared for return to activity by the end of second week. Lastly, most concussed subjects in Subtype 1 were cleared during the third week after injury. This is the highest rate for week three compared to the other four subtypes. Overall differences on RTA days across subtypes were statistically significant at a 95% confidence level (p-vahie=0.033, Kruskal -Wallis H-test of independent samples). When analyzed by pairs, statistical differences were in accordance with the weekly values displayed in FIG. 8A, with Subtype 2 versus 3 reaching a significant - value of 0.004, as well as Subtypes 3 versus 4 with a -value of 0.027. All other comparisons reported p-values higher than a 0.05 significance level.

[0104] FIG. 8B depicts the cumulative distribution for return-to-activity events for each concussion subtype. The differences in rates of recovery shown in FIG. 8A can be observed in this figure. Overall, the fastest rate of recovery is seen in Subtype 3 and the slowest rate of recovery in Subtype 2. For example, more than 50% of Subtype 2, but >70% of Subtype 3, reach RTA by 21 days, with the other subtypes falling in between.

[0105] The relation to symptom burden at the time of injury comparisons were made between conventional measures of self-report symptom burden, using the CSI-22 and the total SAC score. No significant differences were seen between subtype membership and these clinical symptoms, with the median (and IQR) total CSI-22 by subtype 23.5 (29), 23.5 (36.75), 23.0 (27), 24.0 (27.5), 20.0 (33), respectively and total SAC values by subtype 26 (4), 24 (8), 26 (4), 25 (6.75), 26 (5). This suggests that such clinical characterization does not reflect the reported electrophysiological subtypes.

[0106] EXEMPLARY DEVICE FOR IDENTIFYING CONCUSSION SUBTYPES

[0107] Concussion subtypes can be identified, consistent with disclosed embodiments, using devices configured to acquiring EEG signals from patients. As may be appreciated, conventional EEG devices may be bulky, complicated, and intended to use by a skilled technician in a clinical setting. This makes the conventional EEG equipment unfeasible for neuro-triage applications in emergency rooms or at other point-of-care settings. More importantly, the current technology is not field-portable which makes it unfeasible for various field applications, e.g., at a battlefield, or a sports event. [0108] The disclosed systems and methods can be practiced using a field-portable EEG device designed and adapted for use in emergency rooms or at other point-of-care settings. Such a device may be able to perform concussion subtype identification as described herein without requiring the assistance of a skilled technician. Combined with the systems and methods of concussion subtype identification described herein, such as device may provide an improved ability to perform neuro-triage applications over conventional EEG systems.

[0109] FIG. 9 depicts an exemplary schematic of a device 900 for identifying concussion subtypes, consistent with disclosed embodiments. Device 900 can be configured to acquire and process brain electrical signals and provide an assessment of concussion subtype. In an exemplary embodiment, the device 900 can be implemented as a portable device for point-of- care applications. Device 900 can include patient sensor 940, which may be coupled to a base unit 942, which can be handheld, as illustrated in FIG. 9. Patient sensor 940 may include an electrode array 935 comprising at least one disposable neurological electrode to be attached to a patient's head to acquire brain electrical signals. In some embodiments, the electrodes can be configured for sensing both spontaneous brain activity as well as evoked potentials generated in response to applied audio stimuli. In some embodiments, the device 900 can include five (active) channels and three reference channels. The electrode array 935 can include anterior (frontal) electrodes: Fl, F2, F7, F8, AFz (also referred to as Fz') and Fpz (reference electrode) to be attached to a subject's forehead, and electrodes Al and A2 to be placed on the front or back side of the ear lobes, or on the mastoids, in accordance with the International 10/20 electrode placement system (with the exception of AFz).

[0110] FIG. 10 depicts an exemplary placement of device 900 on the head of a patient, consistent with disclosed embodiments. The use of a limited number of electrodes can enable rapid and repeatable placement of the electrodes on a subject, which in turn facilitates efficient, and more accurate, patient monitoring. Further, in some embodiments, the electrodes may be positioned on a low-cost, disposable platform, which can serve as a "one-size-fits-all" sensor. For example, electrodes 935 may be positioned on a head gear that is configured for easy and/or rapid placement on a patient. The head gear may be single-use or disposable. As may be appreciated, the disclosed embodiments are not limited to embodiments using device 900 or the electrode configuration displayed in FIG. 10.

[0111] Consistent with disclosed embodiments, patient sensor 940 can include at least two reusable patient interface cables. These cables can be designed to plug into the base unit 942 and provide direct communication between the patient sensor 940 and the base unit 942. The first cable can be an electrical signal cable 941a, which can be equipped with standard snap connectors to attach to the disposable electrodes placed on the patient's scalp. The second cable can be cable 941b, which can provide stimuli! signals for evoking a response (e.g., mid-latency, late auditory responses, P300, or other suitable responses) from the patient (e.g., cable 941b can provide a connection to the earphone 931 for auditory stimulus delivery, or the like).

[0112] Consistent with disclosed embodiments, the base unit 942 can include an analog electronics module 930, a digital electronics module 950, user interface 946, stimulus generator 954 and battery 944 as illustrated in FIG. 9. In some embodiments, the analog electronics module can receive signals from one or more of the neurological electrodes operatively connected through the electrical cable 941a. The analog module can be configured to amplify, filter, and preprocess the analog waveforms acquired from the electrode channels. The analog module may comprise signal amplifier channels including at least one preamplifier, at least one differential amplifier, at least one common mode detector, and at least one gain stage with filter. The analog amplifier channels correspond to the number of electrode channels. In some embodiments, there are 8 analog amplifier channels corresponding to 8 electrode channels (5 active, 3 reference channels). The analog module 930 may further include a multiplexer (MUX), which combines many analog input signals and outputs that into a single channel, and an analog-to-digital converter (ADC) to digitize the received analog signal. Digital electronics module 950 can then process the digitized data acquired through analog module 930 and can perform analysis of data to aid in interpretation of data pertaining to brain electrical activity. Further, as shown in FIG. 9, the digital electronics module 950 can be operatively connected with a number of additional device components.

[0113] In some embodiments, the analog electronics module 930 can be further configured to check an impedance by feeding a signal back into each electrode. Checking an impedance may function to characterize the effectiveness of connection of a surface electrode to a subject. Such checking can enable a user to test the applied electrodes at a patient site before signal acquisition is started, and also monitor the electrode impedance continuously in real-time throughout the test. In an exemplary embodiment, the impedance of the applied electrodes can be measured periodically, and the impedance values can be displayed on the user interface 946 using a color- coded electrode visual display, which allows the user to monitor the electrode impedances before and during a test session. If an impedance value is found to be unacceptable at the time of the measurement, a warning indication may be displayed on the screen instructing the user to check the electrode connection.

[0114] The digital electronics module 950 comprises a digital signal processor (DSP) 951 for processing the data corresponding to the acquired brain electrical signals, and a memory 952 which stores the instructions for processing the data, such as a DSP algorithm. In some embodiments, device 900 can be configured to use module 950 in performing method 220, depicted in FIG. 2B, or another similar method. For example, when configured with a suitable machine-learning model (e.g., stored in memory 952, or another suitable location), digital electronics module 950 can use DSP 951 in: a. Automatic identification and removal of several types of signal artifacts from the acquired spontaneous brain electrical signal data of the patient; b. Extraction of features used by the machine-learning model; and c. Identification of a concussion subtype for the patient, as described herein.

[0115] In some embodiments, the DSP 951 can be further configured to process auditory evoked potential signals, or the like. For example, in some embodiments, processor 951 can be configured to reconstruct acquired ABR waveforms, remove epochs containing artifacts, filtering, synchronized averaging and computation of Fsp, which is a measure of reconstructed signal quality. Similarly, in some embodiments, processor 951 is configured to process other auditory evoked potentials.

[0116] Consistent with disclosed embodiments, DSP 951 can be configured to implement a DSP algorithm to identify data that is contaminated by non-brain-generated artifacts, such as eye movements, electromyographic activity (EMG) produced by muscle tension, spike (impulse), external noise, etc., as well as unusual electrical activity of the brain not part of the estimation of stationary background state. In some non-limiting embodiments, artifact identification is performed using as input the signals from the five active leads Fpl, Fp2, F7, F8, AFz referenced to linked ears (Al+A2)/2, and sampled at 100 Hz. In some such embodiments, incoming data epochs of 2.56 seconds (256 samples per epoch) can be split into 8 basic data units (sub-epochs) of length 320 ms (32 data points per sub-epoch). Accordingly, in such embodiments, artifact identification can be done on a per-sub-epoch basis and guard bands can be implemented around identified artifact segments of each type. Artifact-free epochs can then be constructed from at most two continuous data segments, with each data segment being no shorter than 960 ms (which corresponds to the time span of 3 contiguous sub-epochs). The resulting artifact-free data can then be processed to extract signal features and classify the extracted features to provide a triage result. As may be appreciated, the disclosed embodiments are not limited to such methods of artifact removal or suppression - other suitable methods known in the art may be used.

[0117] Consistent with disclosed embodiments, DSP 951 can be configured to execute instructions contained in memory 52 to perform an algorithm for quantitative feature extraction from processed signals. In one embodiment, the algorithm extracts various quantitative features from the brain wave frequency bands: Delta (1.5-3.5 Hz), Theta (3.5-7.5 Hz), Alpha (7.5-12.5 Hz), Alphal (7.5-10 Hz), Alpha2 (10-12.5 Hz), Beta (12.5-25 Hz), Beta2 (25-35 Hz), Gamma (35-50 Hz), and high frequency EEG (>50 Hz). In some embodiments, the features computed are absolute power, relative power, mean frequency, coherence, symmetry, fractal dimension, wavelet features, and several statistical harmonics variables. In some embodiments, the features can be include at least one of signal power (e.g., absolute, relative, and the like), mean frequency, connectivity (e.g. asymmetry, coherence, phase lag, phase synchrony, or the like), complexity (e g., fractal dimension and scale-free activity, or the like), information theory (e.g., entropy or the like), or the like.

[0118] Consistent with disclosed embodiments, the feature extraction algorithm takes as input a number of "artifact-free" or "denoised" epochs. In some embodiments, DSP 951 can be configured to perform a linear feature extraction algorithm based on Fast Fourier Transform (FFT). In another embodiment, DSP 951 can be configured to perform a non-linear feature extraction algorithm based on wavelet transforms, such as Discrete Wavelet Transform (DWT), Complex Wavelet Transforms (CWT), Biorthogonal Discrete Wavelet Transform (BDWT), Wavelet Packet Decomposition, etc. A full set of monopolar and bipolar features can be calculated and then transformed for Gaussianity. Once a Gaussian distribution has been demonstrated and age regression applied, statistical Z transformation is performed to produce Z- scores. As may be appreciated, the disclosed embodiments are not limited to such methods of feature extraction - other suitable methods known in the art may be used.

[0119] In some embodiments, device 900 (e.g., using digital electronics module 950, or the like) can be configured to determine a concussion subtype, as described herein. In some embodiments, device 900 can be configured to indicate one or more features making the largest contribution to this determination. In some embodiments, device 900 can be configured to output etiology or type of injury, prognosis, potential sequelae, or the like.

[0120] In some embodiments, device 900 can collect additional brain information, such as auditory evoked potential response data. For example, in some embodiments, the device collects auditory brainstem response (ABR) data and displays the averaged ABR waveforms. For each of the two modalities ("Left ABR," and "Right ABR"), raw data is collected for approximately 2.5 minutes (corresponding to 4096 raw ABR epochs). The ABR waveform is constructed and displayed for lead AFz only, using contra-lateral referencing, which means that for the "Left ABR" modality where the acoustic stimulus is in the left ear, the device computes and displays the ABR waveform for the signal AFz - A2 (by synchronized averaging of artifact-free epochs). Similarly, for the Right ABR modality, the waveform for AFz - Al is computed and displayed. At the end of the ABR data acquisition process, the device computes and displays the Fsp next to the waveform. For the computation of the ABR waveform the following processing steps are performed: bandpass-filtering of raw ABR epochs, rejection of artifacted ("over-range") epochs, followed by Bayesian averaging of the remaining artifact-free epochs. Optionally, adaptive filtering may be performed for ABR waveform reconstruction. As may be appreciated, such information may be used in determine concussion subtype, or in other determinations of brain function. The disclosed embodiments are not limited to embodiments that acquire or use auditory evoked potential response data.

[0121] Consistent with disclosed embodiments, memory 952 can be configured to contain interactive instructions for using and operating device 900 to be displayed on a screen of the user interface 946. The instructions may comprise an interactive feature-rich presentation including a multimedia recording providing audio/video instructions for operating the device, or alternatively simple text, displayed on the screen, illustrating step-by-step instructions for operating and using the device. The inclusion of interactive instructions with the device eliminates the need for extensive training for use, allowing for deployment and use by persons other than medical professionals. Memory 952 may also contain a database, which includes the collected normative data and reference data used for feature classification. In an exemplary embodiment, the database may be accessed from a remote storage device via a wireless or a wired connection. Similarly, data collected from the patient by device 900 may be recorded in the database for future reference.

[0122] Device 900 can be a standalone system or can operate in conjunction with a mobile or stationary device to facilitate display or storage of data, and to signal healthcare personnel when therapeutic action is needed, thereby facilitating early recognition of emergency conditions. Mobile devices can include, but are not limited to, handheld devices and wireless devices distant from, and in communication with, the neuro-triage device. Stationary devices can include, but are not limited to, desktop computers, printers and other peripherals that display or store the results of the neurological evaluation. In an exemplary embodiment, device 900 can be configured to store each patient file, which includes a summary of the session and test results, on a removable non-transitory computer-readable medium 947 (e.g., a memory card, such as compact flash (CF) card, flash memory drive, or the like, or another suitable removable non- transitory computer-readable medium). The user can then use the medium 947 to transfer patient information and procedural data to a computer, or to produce a printout of the data and session summary. In some embodiments, results from the DSP 951 can be transferred directly to an external mobile or stationary device to facilitate display or storage of data. For example, the results from the DSP 951 may be displayed or stored on a computer 948 connected to the base unit 942 using a wired or wireless connection (e.g., a PC interface, such as an USB port, IRDA port, BLUETOOTH® or another wireless link). In some embodiments, the results can be transmitted wirelessly or via a cable to a printer 949 that prints the results to be used by attending medical personnel. In some embodiments, user interface 946 may be configured to communicate patient information and/or procedural data to an attending medical personnel, such as an ER physician, a triage nurse, or an emergency response technician. Information that is conveyed through user interface 946 can include a variety of different data types, including, but not limited to, diagnostic results, intermediate analysis results, usage settings, etc. In some embodiments, the user interface 946 can display the brain electrical signal graphs drawn in real-time for the five active and the three reference channels. In some embodiments configured to acquire auditory evoked potential data, the user interface 946 can display the Right ABR waveform and Left ABR waveform graphs, Fsp value and the actual number of clean epochs used for the computation of the ABR waveforms.

[0123] In some embodiments, user interface 946 can be configured to receive and display usage setting information, such as the name, age and/or other statistics pertaining to the patient. The user interface 946 can include a touchscreen interface for entering the user input. A virtual keypad may be provided on the touchscreen interface for input of patient record fields. Additionally, user interface 946 can be configured to display the battery charge status, available memory status of non-transitory computer-readable medium 947, and electrode impedance values. Further, the neuro-assessment device can transmit data to another mobile or stationary device to facilitate more complex data processing or analysis. For example, the neuro-assessment device, operating in conjunction with computer 948, can send data to be further processed by computer 948. In some embodiments, device 900 can transmit the raw, unprocessed signal acquired from a subject to computer 948 for analyzing the recorded data and outputting the results. The unprocessed brain electrical signals recorded from a subject may also be stored in a remote database for future reference and/or additional signal processing.

[0124] In some embodiments, base unit 942 can include a stimulus generator 954, which is operatively coupled to DSP 951, for applying auditory stimuli to the subject to elicit ABR waveforms. The stimulus generator 954 can interface with earphone 931 positioned in proximity to the patient's ear to deliver auditory stimuli that can generate evoked potentials. DSP 951 can then remove artifacts from the received evoked potential signal and displays the artifact-free waveforms. As may be appreciated, the disclosed embodiments are not limited to embodiments that include stimulus generator 954 or earphone 931, or to embodiments that provide stimuli and record evoked responses to such stimuli. In some embodiments, device 900 can be configured to determine concussion subtypes without providing stimuli or recording evoked responses to such stimuli.

[0125] Additionally, base unit 942 can contain an internal rechargeable battery 944 that can be charged during or in between uses by battery charger 939 connected to an AC outlet 937. [0126] Consistent with disclosed embodiments, device 900 can be used for near-patient testing in emergency rooms, ambulatory setting, and other field applications. In some instances, device 900 can be used in conjunction with CT scan, MRI or other imaging studies to provide complementary or corroborative information about a patient's neurological condition. Device 900 can be used at point-of-care to more provide an assessment of concussion subtype, so that appropriate treatment can be quickly provided, leading to an improved overall clinical outcome. For example, device 900 may be used by an EMT, ER nurse, or any other medical professional during an initial patient processing in the ER or ambulatory setting, which will assist in identifying a concussion subtype of a patient. It will also help ER physicians in corroborating an immediate course of action, prioritizing patients for imaging, or determining if immediate referral to a neurologist or neurosurgeon is required. This in turn will also enable ER personnel to optimize the utilization of resources (e g., physicians' time, use of imaging tests, neuro consults, etc.) in order to provide safe and immediate care to all patients.

[0127] In addition, device 900 is designed to be field-portable, that is, it can be used in locations far removed from a full-service clinic-for example, in remote battlefield situations distant from military healthcare systems, during sporting events for identifying if an injured athlete should be transported for emergency treatment, or at any other remote location where there is limited access to well-trained medical technicians.

[0128] Implementations within the scope of the present disclosure can be partially or entirely realized using a non-transitory, computer-readable storage medium (or multiple such non- transitory, computer-readable storage media of one or more types) containing instructions.

[0129] The non-transitory, computer-readable storage medium can be any storage medium that can be read, written, or otherwise accessed by a general purpose or special purpose computing device, including any processing electronics and/or processing circuitry capable of executing instructions. For example, without limitation, the computer-readable medium can include any volatile semiconductor memory, such as RAM, DRAM, SRAM, T-RAM, Z-RAM, and TTRAM. The computer-readable medium also can include any non-volatile semiconductor memory, such as ROM, PROM, EPROM, EEPROM, NVRAM, flash, nvSRAM, FeRAM, FeTRAM, MRAM, PRAM, CBRAM, SONOS, RRAM, NRAM, racetrack memory, FJG, and Millipede memory. [0130] Further, the computer-readable storage medium can include any non-semiconductor memory, such as optical disk storage, magnetic disk storage, magnetic tape, other magnetic storage devices, or any other medium capable of storing one or more instructions. In one or more implementations, the tangible computer-readable storage medium can be directly coupled to a computing device, while in other implementations, the tangible computer-readable storage medium can be indirectly coupled to a computing device, e.g., via one or more wired connections, one or more wireless connections, or any combination thereof.

[0131] The instructions contained in the non-transitory, computer-readable storage medium can be directly executable or usable to develop executable instructions. For example, such instructions can be realized as executable or non-executable machine code or as instructions in a high-level language that can be compiled to produce executable or non-executable machine code. Further, instructions also can be realized as or can include data. Computer-executable instructions also can be organized in any format, including routines, subroutines, programs, data structures, objects, modules, applications, applets, functions, etc. As recognized by those of skill in the art, details including, but not limited to, the number, structure, sequence, and organization of instructions can vary significantly without varying the underlying logic, function, processing, and output. [0132] The disclosed embodiments can be implemented using microprocessor or multi-core processors that execute software, integrated circuits, such as ASICs or FPGAs, or the like. In one or more implementations, integrated circuits can execute instructions that are stored on the circuit itself.

[0133] Those of skill in the art would appreciate that the various illustrative blocks, modules, elements, components, methods, and algorithms described herein may be implemented as electronic hardware, computer software, or combinations of both. To illustrate this interchangeability of hardware and software, various illustrative blocks, modules, elements, components, methods, and algorithms have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application. Various components and blocks may be arranged differently (e g., arranged in a different order, or partitioned in a different way) all without departing from the scope of the subject technology. [0134] The embodiments may be further described using the following clauses:

[0135] 1. A device for identifying a concussion subtype of a suspected concussion patient, comprising: a patient sensor comprising an electrode array including at least one disposable neurological electrode configured to acquire brain electrical signals when suitably attached to the head of a suspected concussion patient; a handheld base unit operatively connected to the patient sensor, wherein the base unit comprises at least one processor configured to perform operations for identifying a concussion subtype: obtaining biosignals acquired from the suspected concussion patient, the biosignals associated with brain activity of the suspected concussion patient; generating a feature set using the biosignals; generating a concussion subtype indication by applying the feature set to a machine learning model trained to identify concussion subtypes using a training dataset, the training dataset comprising observations and corresponding labels, the corresponding labels associated with the observations using unsupervised clustering of the observations; and providing the concussion subtype indication to support diagnosis or treatment planning for the suspected concussion patient.

[0136] 2. The device of clause 1, wherein: the biosignals comprise EEG signals.

[0137] 3. The device of any one of clauses 1 to 2, wherein: the feature set comprises at least one signal power, mean frequency, connectivity, complexity, or information theory quantitative EEG features.

[0138] 4. The device of any one of clauses 1 to 3, wherein: the machine learning model comprises a logistic regression classifier or a support vector machine classifier.

[0139] 5. The device of any one of clauses 1 to 4, wherein: the concussion subtype indication is associated with at least one of an etiology; clinical sign or symptom; recovery time, outcome, or trajectory; or sequela.

[0140] 6. A method performed using a device for identifying a concussion subtype of a suspected concussion patient, the device including a patient sensor comprising an electrode array including at least one disposable neurological electrode configured to acquire brain electrical signals when suitably attached to the head of the suspected concussion patient, the device further including a base unit operatively connected to the patient sensor, wherein the base unit comprises at least one processor configured to perform operations for identifying a concussion subtype, comprising: acquiring biosignals associated with brain activity from the suspected concussion patient; generating a feature set using the biosignals; associating a concussion subtype with the suspected concussion patient by applying the feature set to a machine learning model trained to identify concussion subtypes using a training dataset, the training dataset comprising observations and corresponding labels, the corresponding labels associated with the observations using unsupervised clustering of the observations; and providing medical information associated with the concussion subtype to support diagnosis or treatment planning for the suspected concussion patient.

[0141] 7. The method of clause 6, wherein: the biosignals comprise EEG signals.

[0142] 8. The method of any one of clauses 6 to 7, wherein: the feature set comprises at least one signal power, mean frequency, connectivity, complexity, or information theory quantitative EEG features.

[0143] 9. The method of any one of clauses 6 to 8, wherein: the machine learning model comprises a logistic regression classifier or a support vector machine classifier.

[0144] 10. The method of any one of clauses 6 to 9, wherein: the medical information comprises at least one of an etiology; clinical feature; recovery time, outcome, or trajectory; or potential sequela.

[0145] 11. The method of any one of clauses 6 to 10, wherein: the medical information comprises at least one of visual disturbance recovery time, outcome, or trajectory; or an indication that the suspected concussion patient possesses an elevated risk of a mental health condition.

[0146] 12. A training system comprising: at least one processor; and at least one non-transitory computer-readable medium containing instructions that, when executed by the at least one processor, cause the training system to perform operations comprising: obtaining biosignal data, the biosignal data acquired from concussion patients and associated with brain activities of the concussion patients; generating observations using the biosignal data, generating a first observation for a first concussion patient comprising extracting first features from first biosignal data for the first concussion patient; generating a training dataset including the observations and associated class labels, each class label corresponding to a concussion subtype; training a machine learning model to predict a class label from an observation; and providing the trained machine learning model.

[0147] 13. The training system of clause 12, wherein: providing the trained machine learning model comprises provisioning or configuring a user device to use the machine learning model; or making the machine learning model available to respond to queries received from another system.

[0148] 14. The training system of any one of clauses 12 to 13, wherein: the biosignal data comprises EEG signals.

[0149] 15. The training system of any one of clauses 12 to 14, wherein: the first features include at least one signal power, mean frequency, connectivity, complexity, or information theory quantitative EEG feature.

[0150] 16. The training system of any one of clauses 12 to 15, wherein: generating the training dataset includes: clustering the observations in groups using unsupervised clustering; and associating a class label for each group with the observations in the group.

[0151] 17. The training system of clause 16, wherein: the unsupervised clustering comprises spectral biclustering.

[0152] 18. The training system of any one of clauses 16 to 17, wherein: generating the training dataset further includes: performing recursive feature elimination using the observations and the associated class labels. [0153] 19. The training system of clause 18, wherein: performing the recursive feature elimination comprises: training a ranking classifier using the observations and the associated class labels, ranking input features to the ranking classifier, and discarding a lowest-ranked input feature.

[0154] 20. The training system of clause 19, wherein: the ranking classifier or the machine learning model comprises a logistic regression classifier or a support vector machine classifier. [0155] As used herein, the phrase “at least one of’ preceding a series of items, with the term “and” or “or” to separate any of the items, modifies the list as a whole, rather than each member of the list (i.e., each item). The phrase “at least one of’ does not require selection of at least one of each item listed; rather, the phrase allows a meaning that includes at least one of any one of the items, and/or at least one of any combination of the items, and/or at least one of each of the items. By way of example, the phrases “at least one of A, B, and C” or “at least one of A, B, or C” each refer to only A, only B, or only C; any combination of A, B, and C; and/or at least one of each of A, B, and C.

Claims

What is claimed is:

1. A device for identifying a concussion subtype of a suspected concussion patient, comprising: a patient sensor comprising an electrode array including at least one disposable neurological electrode configured to acquire brain electrical signals when suitably attached to the head of a suspected concussion patient; a handheld base unit operatively connected to the patient sensor, wherein the base unit comprises at least one processor configured to perform operations for identifying a concussion subtype: obtaining biosignals acquired from the suspected concussion patient, the biosignals associated with brain activity of the suspected concussion patient; generating a feature set using the biosignals; generating a concussion subtype indication by applying the feature set to a machine learning model trained to identify concussion subtypes using a training dataset, the training dataset comprising observations and corresponding labels, the corresponding labels associated with the observations using unsupervised clustering of the observations; and providing the concussion subtype indication to support diagnosis or treatment planning for the suspected concussion patient.

2. The device of claim 1, wherein: the biosignals comprise EEG signals.

3. The device of claim 1, wherein: the feature set comprises at least one signal power, mean frequency, connectivity, complexity, or information theory quantitative EEG features.

4. The device of claim 1, wherein: the machine learning model comprises a logistic regression classifier or a support vector machine classifier.

5. The device of claim 1, wherein: the concussion subtype indication is associated with at least one of an etiology; clinical sign or symptom; recovery time, outcome, or trajectory; or sequela.

6. A method performed using a device for identifying a concussion subtype of a suspected concussion patient, the device including a patient sensor comprising an electrode array including at least one disposable neurological electrode configured to acquire brain electrical signals when suitably attached to the head of the suspected concussion patient, the device further including a base unit operatively connected to the patient sensor, wherein the base unit comprises at least one processor configured to perform operations for identifying a concussion subtype, comprising: acquiring biosignals associated with brain activity from the suspected concussion patient; generating a feature set using the biosignals; associating a concussion subtype with the suspected concussion patient by applying the feature set to a machine learning model trained to identify concussion subtypes using a training dataset, the training dataset comprising observations and corresponding labels, the corresponding labels associated with the observations using unsupervised clustering of the observations; and providing medical information associated with the concussion subtype to support diagnosis or treatment planning for the suspected concussion patient.

7. The method of claim 6, wherein: the biosignals comprise EEG signals.

8. The method of claim 6, wherein: the feature set comprises at least one signal power, mean frequency, connectivity, complexity, or information theory quantitative EEG features.

9. The method of claim 6, wherein: the machine learning model comprises a logistic regression classifier or a support vector machine classifier.

10. The method of claim 6, wherein: the medical information comprises at least one of an etiology; clinical feature; recovery time, outcome, or trajectory; or potential sequela.

11. The method of claim 6, wherein: the medical information comprises at least one of visual disturbance recovery time, outcome, or trajectory; or an indication that the suspected concussion patient possesses an elevated risk of a mental health condition.

12. A training system comprising: at least one processor; and at least one non-transitory computer-readable medium containing instructions that, when executed by the at least one processor, cause the training system to perform operations comprising: obtaining biosignal data, the biosignal data acquired from concussion patients and associated with brain activities of the concussion patients; generating observations using the biosignal data, generating a first observation for a first concussion patient comprising extracting first features from first biosignal data for the first concussion patient; generating a training dataset including the observations and associated class labels, each class label corresponding to a concussion subtype; training a machine learning model to predict a class label from an observation; and providing the trained machine learning model.

13. The training system of claim 12, wherein: providing the trained machine learning model comprises provisioning or configuring a user device to use the machine learning model; or making the machine learning model available to respond to queries received from another system.

14. The training system of claim 12, wherein: the biosignal data comprises EEG signals.

15. The training system of claim 12, wherein: the first features include at least one signal power, mean frequency, connectivity, complexity, or information theory quantitative EEG feature.

16. The training system of claim 12, wherein: generating the training dataset includes: clustering the observations in groups using unsupervised clustering; and associating a class label for each group with the observations in the group.

17. The training system of claim 16, wherein: the unsupervised clustering comprises spectral biclustering.

18. The training system of claim 16, wherein: generating the training dataset further includes: performing recursive feature elimination using the observations and the associated class labels.

19. The training system of claim 18, wherein: performing the recursive feature elimination comprises: training a ranking classifier using the observations and the associated class labels, ranking input features to the ranking classifier, and discarding a lowest-ranked input feature.

20. The training system of claim 19, wherein: the ranking classifier or the machine learning model comprises a logistic regression classifier or a support vector machine classifier.