JPH06505886A - Blood pressure determination method and device - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention] Blood pressure determination method and device Background of the invention 1. Field of invention The present invention uses neural networks to indirectly assess variable physiological parameters such as blood pressure. The present invention relates to an apparatus and a method for doing so.

In particular, the present invention trains a neural network to generate vibrations based on vibration waveforms generated from external blood pressure. The present invention relates to a method for recognizing and interpolating blood pressure values.

2. Conventional technology Blood pressure is one of the main physiological measurements used to determine the status of a patient's cardiovascular system. It is one. During emergency treatment in the operating room or intensive care unit, blood pressure measurements are It is routinely used to monitor and manage the condition of people.

Non-invasive blood pressure assessment methods are generally non-traumatic and do not put the patient at risk. An alternative to invasive methods that require catheters or needles to be inserted into arteries often used in The main drawback of non-invasive methods is that the actual It does not exactly match the blood pressure of the patient. Furthermore, among various non-invasive methods If there is a problem with a poor match, any particular reading obtained by a non-invasive method uncertainty becomes even higher.

Vibrometry has become the most common method used in automated non-invasive blood pressure monitoring methods. There is. The number of automatic vibration measurement non-invasive blood pressure devices in the United States alone easily exceeds 150,000. It seems to be. The advantage over vibrometry or other non-invasive methods is that diastolic and Moreover, not only the systolic pressure but also the mean pressure can be evaluated. Conventional vibration measurement blood pressure monitor Niters use a re-inflated air-filled occlusion cuff that is placed around the limb, usually the upper arm. are doing. The cuff pressure corresponding to the intra-arterial pulse in the artery below the cuff is lower than the diastolic pressure. The small oscillations are recorded as the pressure increases from a low value to a value higher than the systolic pressure. Vibration measurement As a characteristic of a constant waveform, cuff pressure oscillations initially increase in amplitude with increasing cuff pressure.

Although oscillometric methods have become the most dominant non-invasive blood pressure monitoring system, oscillometric waves The origin of the shape and its waveform of blood pressure identified as diastolic, mean and systolic pressure A general theoretical elucidation has not been made regarding the relationship with each attribute. Therefore,( i) an empirical algorithm for assessing mean blood pressure and (2) diastolic and systolic A fixed amplitude algorithm has been introduced to assess systolic blood pressure.

For example, the minimum cuff pressure at which the oscillations reach their maximum value can be used to assess the appropriate mean blood pressure. It is believed that However, the maximum amplitude criterion clearly exceeds the true mean blood pressure. The accuracy is dependent on factors such as the intra-arterial pulse pressure amplitude. death Therefore, ideal measurements are made with the maximum amplitude criterion under all conditions. Not only.

Fixed ratio amplitude standards are used in commercial blood pressure monitors to assess systolic and diastolic pressures. It is used. Such a fixed-ratio amplitude criterion reduces vibration by a fixed amount from the maximum value, e.g. Cuff pressure is confirmed when decreasing by 50 or 80%. Again, fixed ratio amplitude The standard is not constant and changes with blood pressure. The accuracy of such fixed ratio methods is It depends on the empirical percentage and has not been explained theoretically.

In summary, the amplitude oscillometric measurements processed by traditional algorithms represent Blood pressure monitoring is based on minimum cuff pressure versus maximum oscillation for mean blood pressure. For the evaluation of pressure, an empirical percentage-based relationship under a fixed ratio amplitude criterion is used. It's just generalization. Comparing traditional non-invasive measurement techniques to invasive measurements of blood pressure It was found that the non-invasive evaluation can vary by as much as 40% from the true value.

There are at least three main factors that limit the performance of conventional vibration measurement algorithms. 1st The main problem is that vibration waveforms are susceptible to various artifacts and noises. . Typical algorithms are artifacts reflected in vibration measurement waveforms, common noise and inability to cope with other fluctuations. In fact, most conventional vibration measurement algorithms Blood pressure remains constant during the recording period, which can last as long as 10 to 30 seconds. Based directly or indirectly on assumptions.

of any blood pressure assessment by handling artifacts and noise that interfere with the characteristic signal. It is clear that the accuracy is reduced. This can be used to measure the blood pressure that occurs during the recording of vibration measurement waveforms. When combined with periodic changes, oscillometric estimates may be inaccurate rather than accurate determinations of blood pressure. It is clear that the accuracy and usefulness will be limited to providing only a summary representation. It is.

Currently, the second factor limiting the application of traditional vibration measurement algorithms is vibration measurement. It is difficult to interpret empirically the relationships between waveforms and arterial blood pressure attributes as uniform percentages. It's a very simplistic habit. In practice this relationship occurs during a 10-30 second reading. This is an oversimplification, as it varies with changes in arterial blood pressure and pulse pressure. Ru. Check the cuff pressure when the vibration is reduced by 50% from the maximum value to determine the best inflation pressure value. At most, a general recommendation can be made that it can be evaluated as follows. series of Develop more sophisticated algorithms to deal with pulse transformation by pressure-volume curves Although some attempts have been made, these algorithms are sensitive to the subtleties of vibration measurement waveforms. It identifies features and is highly susceptible to artifacts and noise. This makes it difficult to implement as a robust and practical format.

The main obstacles limiting the performance of traditional vibration measurement algorithms are the vibration measurement waveform and arterial arises from nonlinear relationships between blood pressure attributes. This non-linear relationship is This is further complicated by the fact that it is not static. For example, vibration measurement waveforms can be used to is significantly influenced by the interaction between the intraarterial pressure pulse and the nonlinear mechanical properties of the artery. Ru. Because these relationships change with activity, age, and disease, vibration measurement waveforms The use of algorithms that attempt to generalize the relationships between Ru.

Typical pressure monitoring systems rely on the shape, rather than the amplitude, of the vibration measurement pulses to measure blood pressure. Evaluation takes place. This method also suffers from the aforementioned problems. In particular, such a method Requires high-fidelity recording of vibration measurement signals and eliminates the effects of signal noise and artifacts. Always easy to accept.

Therefore, (i) theoretical unexplained, (ii) nonlinear relationships and (iii) blood pressure monitoring Vibrometry to overcome the problem of regular occurrence of noise and artifacts during viewing There is a need for new evaluation methods for constant signals and blood pressure measurement waveforms.

Purpose and summary of the invention Collect and process non-invasive oscillometric blood pressure data and process such data to determine arterial dilatation. Provides a method and apparatus for more accurate assessment of systolic, mean and systolic blood pressure This is the object of the present invention.

Measures physiological variables such as blood pressure with accuracy close to direct measurement without direct invasive measurements. It is another object of the present invention to provide an apparatus and method for evaluating parameters. It is.

Parameter values based on comparison of input data and training data held in the neural network Physiology such as blood pressure by using neural networks as a processing system to determine It is another object of the present invention to provide an apparatus and method for evaluating It is true.

Physiological parameters derived from developmental data that have nonlinear relationships with parameter values A component used as part of a neural network to help evaluate and determine data values. It is another aspect of the invention to provide an apparatus and method for generating group training data. It is a purpose.

Physiological parameters are maintained despite artifacts, noise, and other signal-modifying effects. It is yet another object of the present invention to provide a method and apparatus for determining data values. It is.

blood pressure and other physiological measurements without relying on generalization assumptions that reduce the accuracy of the system. It is another object of the present invention to provide an apparatus and method for determining parameters. It is.

These and other objectives may be based on indirect rather than direct measurements of parameters. This is achieved through indirect, quantitative evaluation methods of altered physiological parameters. This method consists of (i ) identify physiological parameters that are quantitatively monitored and evaluated; and (ii) quantitatively variable. Direct quantitative readings that are dependent on physiological parameters but based on direct measurement of the parameters (iii) generate a signal sequence unsuitable for computer system support; The signals are sent to the computer system containing the input nodes of the neural network for processing and results. At least one output signal for the combined input signal is used as an evaluation value of the physiological parameter. and (iv) output the actual truth for the physiological parameters at the same time as the previous step. Determine the value and (V) make adjustments within the neural network to modify the output signal value and repeat the previous step. the true values of the physiological parameters determined in step ii. Contain the input signals, adjustments and true values associated with the generated signal sequence as training data. The relevant input signals are recorded in memory within the computer system and used to train neural networks. Recognizes online input signals and training output signals stored in computer memory said step sufficient to evaluate the physiological parameter value on the basis of association with It consists of steps that are repeated one after another. related to systolic, mean and diastolic blood pressure. Disclosed are specific applications for implementing the method, as well as apparatus for implementing the method. Also, the waveform A neural network that pre-classifies and ignores noise and artifact signals is also disclosed.

Those skilled in the art will understand other objects and features of the present invention by reading the following detailed description together with the accompanying drawings. I think you get the idea.

Drawing description FIG. 1 is a graphical representation of the components of a conventional oscillometric blood pressure monitoring system; Figure 2 shows when the bladder is inflated to approximately 0 to 90 torr (P cuff). Transcripts actually recorded from the superficial temple artery of a patient in the thoracic intensive care unit. The graph of the derivative (di)/dt) of the juicer bladder pressure, Figure 3, is shown in Figure 2. A graph representing the vibration waveform reconstructed from the derivative of the transducer bladder pressure. Figure 4 shows a graphic diagram corresponding to the pressure readings shown in Figures 2 and 3. Diagram showing the conventional vibration measurement amplitude waveform of the spray, FIG. 5 is a graphical representation of the three-layer neural network used in the embodiments disclosed herein. spray, Figure 6 is a graphical representation of the nodal input/output function, Figure 7 is the signal of 4 torr increments. Graphical representation of normalized vibration measurement amplitude resulting from sampling, Figure 8 A training phase and a method for applying neural networks to determine physiological parameters according to the invention. A block diagram showing how to FIG. 9 shows the present invention operative to generate output values for evaluating physiological parameters. Diagram showing the diagnostic (non-training stage) application of Figures 10a, 10b and 10c show the neural circuits generated for the test. Graphical representation of the web learning curve, Figures 11a, 11b and 11c are test animals. 12a, 12b and 1 Figure 2c shows the training data plotted against the number of hierarchical nodes 50, 250 and l.

Graphical representation of neural network performance after 500 passes, Figures 13a, 13b and and Figure 13c show the performance by the standard deviation of the difference between the invasive measurements and the neural network evaluation. a graphical representation similar to FIGS. 12a, 12b and 12c, in which the Figure 14 shows the conventional algorithm process for diastolic, mean and systolic blood pressure. , numerical data comparing the neural network processes and invasive measurements of the present invention and A diagram showing the corresponding graph data, Figure 15 is a graph comparing a clean signal with a superimposed signal containing random noise. Figure 16 shows a neural network block designed to classify and reject artifacts. rock diagram, Description of examples The first is a block of a blood pressure monitoring and processing system constructed according to the principles of the present invention. It is a diagram. This includes the appropriate hardware necessary to pressurize the cuff according to conventional practice. A blood pressure filter 20 adapted by the software is included. This cuff 20 is representative A band configuration that attaches to the patient's extremities or a surface band that attaches to the patient's head. Can be used as an arterial blood pressure monitor. Temple artery, first described here Disclose blood pressure pad. However, those skilled in the art will not be able to apply the principles of the present invention. It seems obvious that the conventional occlusion pressure force can be used instead.

The cuff 20 interfaces with the patient's pulse side to provide non-invasive vibratory blood pressure data. , which inflates the transducer bladder of cuff 20 and increases the internal pressure. A parallel interface is installed in the satellite box 23 containing the necessary hardware for the measurements. Processing is performed within the computer 21 connected via the face card 22.

Software-controlled DC Romega 80 air pump wipes the transducer bladder Used to make a difference. Most vibratory blood pressure monitors have a deflation lamp. I use software to inflate the transducer bladder. A controlled DC Romega 80 air pump is used. Most vibratory blood pressure monitors Although the vibration waveform is recorded using a deflation lamp, superficial temple artery monitoring Inflation lamps are used for the turret. Dampens pump vibration and smooths it out The air pump is used to inflate the transducer bladder with a pressure gradient. The output is passed to two rigid 10m1 volume chambers, each separated by air resistance. . From the damping chamber to the transducer bladder using a manually adjustable needle valve 24 The flow rate of the pump is controlled and can have a variable pressure ramp rate.

The transducer bladder of the superficial temple artery blood pressure pad has an inner diameter of approximately 1.5 mm. Connected to the output of the needle valve 24 via an air line 25 with a length of approximately 1.5 m (mm) has been done. From the output of the needle valve 24 the pressure drainer and software A secondary line is connected to the control solenoid valve.

Pressure transducer is superficial temple artery blood pressure pad transducer bladder Used to record inflation ramp and vibration measurement waveforms from.

Solenoid valve releases pressure in transducer bladder each time blood pressure is determined Acts as a dump valve. With regard to Figure 1, if you are in the same industry, the satellite box Box 23 includes the pump and pneumatic circuit, pressure transducer analog amplifier/filter. router, 12-bit A/D converter and parallel board. I think you get the idea.

This combination hardware responds to cuff 20 inflation requests and initial filtering and processing of the signal. do the same thing. The digital signal then passes through connection line 27 to the computer interface. The information is sent to the chair card 22. Software in the computer 21 allows later data to be Data collection, processing, and data display are controlled.

The vibration measurement waveform is a transducer plotted as a function of transducer bladder pressure. – consists of sub-bladder pressure oscillations and uses the derivative of the transducer bladder pressure signal It is configured by software. The derivative of transducer bladder pressure is Due to pressure changes caused by the inflation lamp and volumetric vibrations transmitted from the lower arteries It is the sum of pressure changes. The purpose of reconstructing the vibration measurement waveform is to reconstruct the derivative corresponding to volumetric vibration. Separate the number signal components from the derivative signal and then beat-by-beat the resulting signal. The goal is to integrate and recover the original pressure oscillation.

Figures 2 and 3 show vibration recordings from the superficial temporal artery. Second The figure shows the case when the transducer bladder is inflated to approximately 0 to 90 torr. The derivative of transducer bladder pressure is shown. Positive offset of derivative signal That is, the bias corresponds to the slope of the inflation ramp. Figure 3 shows the trans Figure 3 shows a vibration measurement waveform consisting of the derivative of the juicer bladder pressure signal.

Note that one oscillation or pulsation starts at and returns to approximately the same diastolic pressure level. If you look closely, the vibration measurement waveform will be further purified. Therefore, - the derivative over the beats The sum of the signals must be zero. Any non-zero sum is considered part of the ramp signal. is subtracted from the derivative signal over the beat period. The adjusted derivative signal is Integrate over the beat period to more accurately reconstruct the oscillations or beats. Reconstruction The resulting vibration measurement waveform is shown in FIG. 4 in a conventional format.

In the conventional technology for evaluating diastolic, mean, and systolic intraarterial blood pressure, Although certain points are identified empirically, the present invention examines the entire waveform. For example, Figure 4 shows the 80% diastolic algorithm, the maximum amplitude averaging algorithm, and the 50% yield algorithm. By applying systolic algorithm software, baseline blood pressure attributes can be evaluated using conventional techniques. can be valued. These points are on the graph shown to approximate blood pressure attributes. It represents an empirical and somewhat arbitrary point.

However, as already pointed out, these differences are due to patient age and physiological differences. This is a generalization that does not represent accurate changes in blood pressure. signal by neural network Processing components and waveforms not only improves the accuracy of blood pressure determination, but also improves the accuracy of blood pressure determination. It also reduces the effects of noise and artifacts that were processed together with the constituent periodic signals. It has been found that this is an effective method for solving the problem.

Neural networks are based on nervous system models and pre-processed using adaptive signal processing techniques. to develop training data that facilitates recognition of previously observed conditions. neural network training provides a means to transform a given input signal into an appropriate output signal. precept Use a trained data set to obtain an appropriate output for a given input in a statistical sense. applied to the various nodes that make up the network until the network is optimized to The applied neural network weights are corrected.

The present invention provides a method for evaluating physiological parameters assessed by indirect measurements on the patient's body. It applies neural networks to identification and evaluation. Such indirect measurements Physiological events are based on the occurrence of signal sequences that quantitatively depend on physiological parameters. Suitable for non-invasive monitoring of elephants. As mentioned above, blood pressure A good example of such a non-invasive evaluation is based on the monitoring of signals generated in dynamic measurement methods. It is.

Training a neural network to convert non-invasive vibration measurement signals into arterial blood pressure estimates Can be done. Rather than identifying a single occurrence, such as the point of maximum vibration, the cycle of the entire vibration measurement signal is Because it uses a network process, the network is inherently more complex than standard vibration measurement algorithms. This makes it more robust (less affected by noise and artifacts). Additionally, blood pressure Unlike standard algorithms whose accuracy varies depending on factors such as can perform nonlinear processing of the input signal and is therefore relatively stable over a wide range of pressures. Be consistent.

Neural networks can be specified by their architecture. This includes nodes numbers and their interconnection relationships, nodal characteristics such as input/output functions, and nodal interconnections during training. Contains learning or training rules that define how to adapt the

The power of neural networks uses nonlinear functions to process nodal inputs and As inputs to many nodes can operate simultaneously on network inputs rather than being constrained to a single node. This is partly caused by the use of emerging parallel distributed processing.

In accordance with the present invention, as shown in FIG. Designed to process waveforms. A three-layer system has an input layer 3o, one solar calendar 3 1 and an output layer 32. Although a single output layer is generally referred to in this disclosure, , multiple output layers when separate distinct output values are expanded from the same output dataset can also be achieved. For example, the present invention provides a method for each set of inputs from a patient. Three outputs can be implemented representing diastolic, mean and systolic blood pressure. However, Therefore, -output does not have a restrictive meaning, but rather refers to at least one output to the network. must be interpreted in that sense.

In Figure 5, each input node shown as P (n), P (n - 1), etc. is a link with a load. 33 to each solar calendar node 34. Similarly, each solar node 34 has a load The output layer is connected to a single output layer node 36 via a link 35.

For the application of evaluating blood pressure attributes, the neural network is provided with 40 input nodes 37. 40 incremental signal samples of normalized vibration measurement amplitude waveform (cuff pressure vibration) dynamic vibration radiation versus cuff pressure). These samples are 20-17 It is withdrawn at uniform intervals of 4 torr over a cuff pressure range of 6 torr. child These 40 input samples are stored in computer memory and then input to the input layer 30. Sent to 0 input nodes simultaneously. That is, if the first sample is sent to P(n) The second sample is sent to P(n-1), and so on. This is sent A total set of sampled signals are simultaneously received at the input layer 30 to perform a single diastolic test procedure. represents a sample of a waveform representing .

This input signal is applied to one or more solar calendars 31 by applying the weighting factors at the interconnected nodes. A nodal quantity relationship between the input signal 38 and the desired output signal 39 is established. Ru.

This process involves determining the specific values of blood pressure and other physiological parameters to be matched. Loading within the neural network that modifies the value of the output signal 39 is performed at link 33.35. Includes adjustments made. This is the input signal when the output value 39 is processed through the network. This is done by a training sequence that results in a known value generated by adjusting No. 38. God The procedure for training the network to obtain this result begins with the weighted Relink 33.35 Establish appropriate weighting factors within a similar input signal set for future diagnostic tests. Loading factor in link 33.35 stored in memory when 38 occurs By applying a neural network or converting such input data into a desired output signal 39 This includes joining with.

Details of this procedure will be described later.

Nodes are typically internal thresholds or offsets and non-linearities that process the nodal inputs. It is characterized by the type of

The internal threshold and offset of the seed layer node 31 are a generalization of the Widlohoff delta rule. is determined by adaptively using a known backpropagation algorithm. Error backpropagation a The algorithm is designed to minimize the mean squared error between the desired and actual output of the network. A designed gradient descent algorithm. Train the network to generate the error term. The dataset used for training includes not only the network input but also the data specified during supervised training. The desired output must also be included.

Application of the backpropagation algorithm consists of the following three steps.

1. Input data is processed forward within the network and output is obtained. desired output and actual An error term is calculated using the difference with the network output.

2. The error term is back-propagated through the network and is smoothed by correcting the nodal phase coupling loads and nodal thresholds. The mean squared error is minimized.

3. Interactive adaptation process adapts new input data to step l and and 2 are repeated. Generally, after a network reaches a certain convergence level, e.g. For example, when the error drops to 10% of the desired output, the adaptation is stopped and the coupling loads are saved. Ru.

Next, the formulas and formulas used to implement the backpropagation algorithm for vibration measurement waveform processing are and a representative list of steps.

1. Initialize the joint load between nodes to an arbitrarily small value.

2. Provide training data to the network (i.e. train the network using input data) Operates in feedforward mode. ) 3. Use the network output and desired output to apply the inter-nodal coupling loads as follows: let

Load update formula %formula%() Wz(n) is the output node from the hidden node i at time n or the hidden node from the input node i Load on point j.

Xi (n) is the output or input at node i.

Gain product, convergence factor.

e, (n) are error terms calculated for node j.

Error term when j is a solar calendar node: e r (n) = [di (n) - y r (n)] [1 - y r (n) Hy r (n)] wh+(n) Error term when k is the output layer node: eh (b) = (dm (n) -)'h (n)) x + (n) is the solar calendar node The output of j, k exceeds the output node.

4. Repeat from step 2.

For each solar calendar node 34, inputs 33.33a, 33b.

The sum of loads 33c and 33d (dot product of nodal input and load coefficient) has the form shown in Figure 6. is passed through the S-shaped nonlinear form of Eq.

It is desirable to continuously read the arterial blood pressure output in torr from the neural network. Therefore, S-shaped nonlinearity is not applied to the output layer. The output layer nodes are loaded with solar calendar nodes. Acts as a simple adder of outputs. In this addition formula, the nodes are either threshold or off. The set and function are S-shaped nonlinear as follows.

■ f (a) = 1 + c - (a - θ) The 40 sample input to the network input node layer 30 is the normalized vibration measurement vibration shown in FIG. It is represented by a width waveform. This amplitude waveform is part of the process of the invention and is and the signal sequence generated from the pressure transducer is 4 torr apart. A set of 40 samples representing the total pressure range covered by the diagnostic test procedure. A loose sample signal is obtained. For each sample signal, the maximum value of that sample signal is Features that constitute large amplitudes are identified. Other diagnostics involving vibration signals with amplitude characteristics This same procedure can be applied for testing. Next, using this feature, Figure 7 A reference waveform is developed, and each sample amplitude value is a point representing the cuff pressure amplitude for the diagnostic test. form a trajectory. This waveform is realized when represented by the 40 input nodes of the neural network. This results in an image or pattern corresponding to the actual blood pressure value. The neural network is divided into this waveform. The output layer recognizes the actual blood pressure value that is attached and converts this value in the output layer when receiving the same set of input signals. trained to generate a value.

Based on the above description and FIG. Common devices that use neural networks to make measurements without invasive activities include: It can be summarized as follows. The device has physiological parameters that are quantitatively monitored and evaluated. It includes sensing means 20 for indirectly detecting changes in the meter. The sensing means is a parameter selected according to the nature of the data and typically already used to perform such evaluations. This is a conventional diagnostic device. As mentioned above, we are currently generating vibration measurement signals. The resulting blood pressure force forms the sensing means for blood pressure applications. Pulse oximetry This is another procedure that can be adapted to processing by the network. In this case, blood oxygen saturation A degree rating is transmitted or reflected from the human tissue. The third application area is generally This is called diluted cardiac output. This procedure involves injecting a dye or hot solution into the vascular system. By processing time-dependent concentration or temperature signals that generate The flow is evaluated. Obviously, different sensing devices are used in the latter two cases and the Quantitatively dependent on a variable physiological parameter but based on direct measurement of that parameter An appropriate signal not suitable for direct quantitative reading is used.

The sensing device and the associated signal generating means 23 jointly apply to the input nodes of the neural network. generate the required set of signals that can be obtained. The neural network is connected to a means of generating data Support computer system 21 operative to control collection, processing and display Contains. If you are in the same industry, the computer system is a separate computer. Can be specifically designed to realize neural networks without using systems Other data processing devices such as hardware analog circuits and integrated circuits may be included. I think you get the idea.

Although neural networks have been described for one embodiment, they generally include (i) developmental procedures; a series of input nodes receiving signals from the stage, (ii) individually coupled to each input node; a series of hidden nodes, (iii) at least one connected to each hidden node to give the desired output value; Contains one output node. The neural network extracts one signal from the signal received at the input node. means for generating an output signal, the output signal providing training evaluation of a physiological parameter; A value is given.

The computer system receives input signals and outputs specified during such training. The instructions voiced within the neural network about the relationship between the desired physiological parameter and the desired value. It also operates as a data storage means for storing training data. computer system is It also displays the evaluation values of physiological parameters based on the output signals from the neural network. It also provides reading means.

When used as part of a training system, the invention includes invasiveness adapted by means of determining the actual true value of the physiological parameter followed by Also included is a detection means. The invasive detection means receives the sample signal from the generating means. applicable at the same time. The computer system has corresponding information that is sent to the neural network. Memory means are provided for storing the true values of the parameters in association with the input signals.

These related values and signals will be used to create a similar set of input signals at the input nodes of the future neural network. It is called and associated when a reoccurrence occurs.

The number of input nodes varies depending on the number of input signals, but is minimal to establish a minimum statistical image. also requires two input nodes. Similarly, hidden nodes also vary in number and level. General A simple solar calendar is appropriate, and usually at least five nodes are input nodes and a single output node. Construct a mono- solar calendar between the points. If additional boundary conditions are required within the neural network, , a multi-year calendar can be applied.

The computer also provides selection control means for sampling the periodic signal from the generating means. do. As shown in the previous example, the diagnostic test procedure consists of 40 samples over a pressure range. Signals are retrieved, stored in memory, and sent to the input nodes of the neural network on a simultaneous basis. sent later. Generally, at least one feature is discernible within these sample signals. This feature is used as a feature signal that has a dependent relationship with physiological parameters. It can be processed by a circuit network.

When using the system of the invention in training mode, computer memory or Some of the other memory measures are weighting factors and mean intra-arterial pressure, systolic intra-arterial pressure and parameters that can be used to generate physiological parameter values including training data including parameter values. Applying the present invention to diagnostic applications invasive detection methods necessary to determine physiological parameter values. There is no need to connect inventions. The device is equipped with neural networks and online data input. All you need to do is include the recorded training data needed to deploy the federation.

In this diagnostic configuration, the device is configured and configured to determine the named blood pressure attribute. and three separate neural networks trained to achieve the same result. It can be a single neural network with three outputs configured as follows. neural circuit Anyone with general skill in the art regarding network systems will be able to follow the teachings of the present invention. Further details about the technical implementation of the network are not considered necessary.

FIG. 8 illustrates the general procedural steps associated with the present invention in a broader context. First step The step identifies physiological parameters to be quantitatively monitored and evaluated. item 41 is the associated physiological parameter of blood pressure and diastolic, mean and systolic blood pressure. represents the data. Cuff 41 is associated with associated hardware that assists in movement of the cuff in a conventional manner. It also represents. The next step quantitatively relies on variable physiological parameters However, the reliability is not suitable for making direct quantitative readings based on direct measurements of parameters. A number sequence 42 is generated. In the third step, these signals are computer system 44 for processing; The coupled input signal includes an input node 45 of a neural network 46 supported by a system 44. 43 provides the ability to generate a single output signal 47 for signal 43. This output signal 47 gives an estimated value of the physiological parameter corresponding to the reference input signal 43.

In the training mode, the device shown in Figure 8 generates a signal simultaneously with the occurrence of the signal shown in item 42. including the step of determining the actual true value of the physical parameter. This procedure is placed invasively within the patient and directly reads the actual blood pressure value and connects the computer via line 50. an intravenous device 49 transmitting data to a data system 44; This true of the parameter The values are processed by the computer system and stored as training data 51. child The value of is transmitted over line 52 as the desired output value 47. Next, within the neural network 46 The value of the signal adjusted by and transmitted from the output node 53 is transmitted from the training data 51. is corrected to be equal to the desired output value 47. Typically, this is the neural gyrus The solar calendar of the input node and the node 54 in the road network, and the phase between the solar calendar of the node 54 and the output node 53 This is done by applying a load factor to the mutually connected nodes. During the training phase, the input signal is and modified to reach an output value equal to the output value produced by the invasive measurement 49. The input signal 42, along with any necessary adjustments and weighting factors, is saved as training data 51. . These data are recorded centrally as training data 51 and no invasive measurements are taken. used for actual online diagnostic measurements of patients.

Repeat this series of steps a sufficient number of times to train the neural network and Based on the association of online input signals, the relevant input signals are recognized and the desired physiological The values of the analytical parameters are evaluated.

As mentioned above, this method uses vibrations that generate waveforms corresponding to a single diagnostic test procedure. Particularly applicable to signals. In this case, this diagnostic test procedure expressed by conventional vibrometry methods that generate the desired signal sequence. like this For use in a typical system, the amplitude or frequency of the vibration signal must be within physiological parameters. It is desirable that the data change depending on the data and dependencies. This allows the true value of the parameter to be Through the actual training that is taught to the neural network in conjunction with the input signal, the neural network is You can learn about various relationships.

Another value of using neural networks is that the entire signal sequence 42 originally generated is Analyze and interpolate from a few sample signals without processing to determine parameter values. It is the ability to generate accurate evaluations. According to this method, the computer system or a signal sequence that can be received directly via line 55 by other types of selection control means. A plurality of sample signals are selected from the source 42. The computer system uses characteristic signals and of the signal amplitude, etc., in the sampled signal that can be processed by the neural network. Identify at least one feature. The normalized vibration measurement amplitude waveform shown in Figure 7 is , 4 to 4 without processing the entire signal as shown in the waveforms of Figures 3 and 4. How can representative waveforms be generated using 40 signals selected in rr increments? It shows. In the present invention, only three sample signals are selected and according to the teachings of the present invention. blood pressure monitoring system to obtain accurate results. was successful.

Obviously, the generation of significant waveforms depends on the training ability of the neural network regarding the desired parameters. At least two sample signals are required to generate the signal. Therefore, processing the analysis And since the associated step can be performed using only a few signals rather than the entire range of specified signals, neural circuits are The efficiency of the road network system will be improved. In general, neural circuits are The minimum number of input nodes required for the road network is determined.

In the example, the specific implementation method includes developing waveforms for each diagnostic test procedure. The rare predetermined number of sample signals approximately corresponds to the number of input nodes in the network. serial message The signal is stored in memory and simultaneously sent to the input node of the neural network as a representative waveform. Sent. The steps of the invention shown in FIG. 8 are specific training for generating blood pressure training data. Application to a session is done in the following specific format: Is it a pressure sensing means? A vibration measurement signal sequence is generated.

This means, indicated by the blood pressure graph 41 in FIG. connected to. Computer system 44 identifies a set of sample signals in predetermined increments. The main features of the vibration measurement signal constitute the pulse amplitude. . Measure the pressure value with the pressure sensing means along with the corresponding pulse amplitude signal mentioned in the previous step. The process is continued by constant recording. Heartbeat, as shown in item 49 An invasive blood pressure measurement is performed simultaneously with the generation of an oscillometric signal representative of . This true value is The data is sent to the computer system and recorded as part of the training data 51. at the same time, A sample feature signal representing the pulse amplitude is sent to input node 45 and is processed by the neural network. It is processed. Invasively adjust the network output signal by applying weighting factors to make appropriate adjustments. Forcibly match the determined output value. A neural network recognizes a set of input signals and This training is repeated until an accurate assessment of the blood pressure value can be determined.

When applied to vibration measurement techniques, the range of measured and recorded pressure values is typically in predetermined increments. It ranges from approximately 20 to 200 torr. In the present invention, a suitable increment is 4 torr. , representing approximately 40-45 sample signals.

The system was tested on five dogs. Arterial diastolic, mean and systolic blood pressure A total of 425 oscillometric amplitude waveform recordings were obtained with simultaneous invasive measurements of (− Approximately 85 hits per record). to assess diastolic, mean and systolic blood pressure respectively Three separate neural networks were used. These systems use the backpropagation algorithm Trained using rhythm.

Training the network using trained on-4/test-on-1 procedures and tested. After training the network with data from four dogs, Stop applying the calendar thresholds and circuit adjustment point coupling weights and recirculate the data from the fifth dog. Arterial diastolic, mean, or systolic blood pressure was evaluated by processing the blood pressure according to the network. Professional Repeat the dog five times using data from each person and data from the other four dogs. Tested on trained network.

Since there is no clear rule to determine the optimal number of hidden nodes, each network The training and testing process was repeated using 5.31.63 solar calendar nodes.

The convergence coefficients appearing in the equations in FIG. 6 were set to o and ooi, respectively. The training data is Each network was adaptively passed a total of 1.500 times. Following each adaptation pass, (− per animal The training data (340 vibration measurement readings from 4 dogs with 85 results) was applied to the neural circuit. The convergence level was evaluated by sequential processing using a network. Test data from the 5th dog (8 5 vibration measurement readings) are then sequentially processed by the circuitry to perform neural circuits at various levels of convergence. The performance of the road network was evaluated. The convergence coefficient and the total number of passes of the training data are determined by the appropriate convergence rate, The final convergence level and steady state oscillations were chosen to obtain.

The convergence level was quantified by mean error, standard deviation of error, and mean squared error.

The error is between the desired value (invasive arterial blood pressure measurement value) and the actual network output (non-invasive evaluation value) It was calculated as the difference between The mean squared error will be minimized by the backpropagation algorithm. This variable is an appropriate measure of the convergence level.

Network performance on test data is always based on arterial measurements of arterial blood pressure and non-invasive neural circuits. The evaluation was performed by converting the average difference and standard deviation of road network evaluation values. Obtain the evaluation value of arterial blood pressure Conventional vibration measurement algorithms were also used to Vibration first reaches maximum value Mean blood pressure was evaluated as the cuff pressure at When the vibration is reduced to 50% of the maximum amplitude Systolic blood pressure was evaluated as cuff pressure. Cuff pressure at which vibration increases to 80% of maximum amplitude Systolic blood pressure was evaluated as follows.

An example of using the present invention in the diagnosis phase rather than the training phase is shown in FIG. nerve Although the processing by the circuitry is essentially the same, the purpose of diagnostic applications is to eliminate the discomfort of invasive techniques. The purpose of assessing such measurements without causing discomfort or trauma is to avoid invasive methods. Simultaneous measurements are not performed. In summary, this quantitative assessment of variable physiological parameters The method identifies the evaluated parameter and generates an online generate a sequence of signal signals and send these signals to the input nodes of the neural network. be exposed. At this stage, the neural network is properly trained and is part of the training data. (based on a comparison with the signal received at the input node) The training data used to identify the closest parameter value is stored in its memory. It has been paid. The actual output value depends on the generation of the output value corresponding to such an input signal. It is obtained by processing online signals within the neural network. If you are in the same industry, the above training As you may find it easier to understand how this diagnostic phase compares to the There seems no need to repeat the above explanation. In fact, the foregoing description is incorporated by reference. The reason for this is that the neural network processes online input data and that the neural network blood pressure and other data based on comparison and interpolation when enabled by training data. This is because it is related to generating the evaluation pressure value of the parameter.

Figures 10a, 10b and 10c show that the network is trained after each adaptation pass. An example of a learning curve obtained by sequentially processing data is shown. This data is based on three solar calendar nodes. It corresponds to a circuit network that exists. Figures 11, 11a, 11b and 11c The figure is obtained by sequentially processing test data using a neural network with 63 solar calendar nodes. The corresponding performance curves are shown. These curves are inside the network, and in the 1.500 times above, This represents the case where the training data is presented a total of s, o o o times. 1st As shown in Figure 0a, Figure 1Ob, and Figure 1OC, the average error learning curve initially increases rapidly. It changes, sometimes changing its sign, and then moving up and down gently and unevenly before heading towards 0.

Both the standard deviation and mean square error of the error are expressed by exponential functions that decay unevenly. It will be done. It is found that the convergence rate decreases as the number of solar calendar nodes increases.

However, the mean error, standard deviation of error, mean squared error, and steady state oscillation also decreases as the number of solar calendar nodes increases.

Average difference and difference performance as shown in Figures 11a, 11b and 11'c. The standard deviation of a curve generally has the same shape as the corresponding learning curve. mean standard deviation The difference is generally between the corresponding learning curves (Figures 10a, 10b and 10c). They have the same shape. However, as training increases, the mean difference approaches 0, but the standard deviation of the difference The difference increases.

Therefore, increasing the convergence level or reducing the mean squared error does not necessarily mean that the test data Good performance for data is not guaranteed. This effect is due to the fact that the network uses training data. It is thought that this is because the film-forming ability is lost as the film-forming ability is lost.

In summary, the convergence pattern varies with the number of solar nodes and the number of passes in the training set. However, all of the different neural network architectures converge well. General Generally speaking, increasing the hidden number of nodes increases the convergence level of the training data and reduces the damping exponential function. Numerical test data performance is improved.

Increasing the number of solar calendar nodes also reduces steady-state oscillations in the learning and performance curves. training Increasing the number of passes through the data increases the level of convergence, but especially after long tests data performance will not necessarily improve. The cost of improved performance is an increase in the number of interconnections. This increases the amount of time it takes to train or process data through the network. be.

Figures 12a, 12b, 12c and 13a, 13b, 13c The figures show invasive measurements of arterial blood pressure and non-invasive nerve measurements plotted against the number of solar calendar nodes, respectively. The difference from the network evaluation value and the standard deviation of the difference are shown. The network performance is 50% of the training data. ．． Test data is processed sequentially by the network after 250 and 1500 adaptation passes. This will allow you to be evaluated at different training levels. Mean difference and standard difference Figure 12a , Fig. 12b, Fig. 12c and standard deviation of the difference Fig. 13a, Fig. 13b, Fig. 13 Both figures decrease as the solar calendar number increases. Performance improvement follows a bumpy decay exponential curve It was said that As mentioned above, increased training does not necessarily guarantee better performance. . Good performance (minimum standard deviation of difference) for assessing diastolic, mean or systolic blood pressure is Using a network with 63 solar nodes (the maximum number of solar nodes tested) achieved. Best performance was achieved after 422 training baths for diastolic evaluation values. , after 18 training passes for the average evaluation value and after 548 training passes for the systolic phase. achieved.

Neural network oscillations compared to traditional algorithms used to determine blood pressure parameters The measuring blood pressure evaluation device showed equivalent or better performance for data obtained from 5 dogs. Ta.

A method for evaluating blood pressure and other physiological parameters using neural networks is vibration measurement amplitude waveforms. can be a powerful alternative to traditional algorithmic processing. Neural networks are Requires detailed knowledge of the relationship between force (oscillometric waveforms) and output (arterial blood pressure attributes). One advantage arises from this. Neural networks use supervised training to adapt inputs appropriately. An internal rule set is developed that is used to transform or map to the output.

Another great advantage of neural networks is that they are very easy to implement. one time When properly trained, neural networks readily respond to the occurrence of appropriate parameter estimates. can be done. Furthermore, neural network systems are It is inherently robust in that it is not sensitive to artifacts or noise. ( typically relies on the identification of a single event (e.g., the lowest cuff pressure at which maximum oscillation occurs) Unlike conventional algorithms, the overall vibration measurement waveform is processed by a neural network to create the desired result. Evaluation values of blood pressure attributes can be obtained.

The favorable results of the present invention when compared with conventional algorithm techniques are summarized in Figure 14. promise Figure 14 shows not only the neural network system of the present invention but also invasive measurement values and A comparison table with conventional algorithm techniques is also disclosed. The first row of the table is the average blood pressure. difference, plus or minus between invasive measurements and non-invasive conventional algorithms The standard deviation and neural network evaluation value are shown. These statistics are based on the test animals mentioned above. Calculated using data generated during

The second row shows the mean and standard deviation for each person, and the measured value of the variation in the target amount. It is intended to improve the The attached graph shows the accuracy of the conventional algorithm as blood pressure increases. This dramatically shows that the accuracy of neural networks remains relatively constant. Ru.

Therefore, the distributed non-linear processing capabilities of the neural network system disclosed herein can be expanded. Significant benefits and potential for maintaining blood pressure assessment accuracy across a wide range of psychological conditions You can get sex.

Neural networks can also be used as part of child classification systems to identify the properties of some input signal. It can also be used. For example, when a set of input signals arrives at the input nodes of a neural network, This makes it easier to detect certain patterns that are unique to children rather than adult patients. Wear. Such child classification applications are generally applied to a broad category called patient-induced states. This is useful in distinguishing between different patient conditions that may occur. fall into another unique classification Age, body size, medical conditions, and other conditions are reproduced at the input nodes of the neural network. It can be detected by the pattern. Once detected, the neural network This can be done by constraining the selected training data to only those that are applicable to the given classification. processing of information can be simplified.

As an example, train a neural network to recognize blood pressure attributes on pediatric patients. be able to. By using child classification terms, neural networks can detect that the input signal has a periodic pattern. training data and input developed specifically for pediatric patients. Comparisons with force data can be limited. for equipment-induced conditions that represent failures. Neural networks can be similarly applied to child classification rules. Common with diagnostic equipment By referring to the training data that enables neural circuits to recognize failure conditions, the attending physician can It is possible to issue warnings for equipment correction and maintenance in a timely manner.

Similarly, the neural network of the present invention is free from noise and artifacts received at its input nodes. It can be trained to recognize input. As mentioned above, this technology is particularly Applied to measurements in test animals. The identification procedure involves human observation of the waveform graph. The vibration waveform quality is determined. This is done by observing the waveform and checking for noise and artifacts. Examine the occurrence of the signal and then rate it as “excellent,” “good,” or “artifact.” This is done by assigning a “quality” factor. Select training samples from a total of 245 waveforms It is processed by a neural network with 60 human nodes, 15 intermediate hidden nodes, and a single output. I understood. This network uses a supervised probabilistic method to It was trained to calculate "quality" numbers at the output nodes. In the actual experiment, The selected numbers are 500 representing an excellent waveform, O representing a good waveform, and 500 representing an excellent waveform. The value was 0, which represents the current state. At the end of training, the network will Low numbers and high numbers representing good and excellent waveforms can be calculated consistently. Ta. Therefore, the worst quality waveform and the good waveform can be distinguished, and the network It was able to identify and reject artifacts and noise signals. This is signal training This was the case for both datasets and non-training datasets. Also, this step is tested on a non-training dataset, the network is tested on the first human classification program. Some waveforms that were misclassified during the process were properly classified.

FIG. 15 is a graph of normal oscillometric pulse amplitude versus cuff pressure. High quality A clear signal is indicated by a small angle point indicator and is free from random noise or artifact signals. The weight is indicated with a + sign. Neural networks are inadequate to process each of these signals. It has been confirmed that it has the ability to identify and reject high-quality signals and record and process high-quality signals. It was.

This functionality of child classification is illustrated in FIG. Again, the selection parameters are indicated by cuff 60. Blood pressure generated using an oscillatory measurement system. In this case, the child classification network is Recognizes signals modified by external noise, classifies them as artifacts, and rejects them. trained to be stored as training data for future recognition . All other signals are recognized as high quality signals 65 and the 63 neural network as described above. The blood pressure is processed and evaluated as an output signal 67.

The development of training data trains the neural network to recognize certain blood pressure parameter values. The procedure is similar to that described for Typically, the signal input can also be modified. Otherwise, it is classified as an artifact and a load factor is applied to mutually connected nodes in the neural network 62. A number is applied and defined as the modified signal and artifacts 63 received at the input node. A nodal quantity relationship is established between the output signal and the desired output signal. These relationships and values are Signal inputs stored in computer memory and of no predetermined quality that will be combined in the future. It will be done. During the training process of the blood pressure neural network, the child classification neural network 62 - identify and discard artifact signals so they are not stored as part of the training data It is useful for This means that all artifacts and noise signals are classified and rejected. It operates to improve the accuracy of stored training data. in this case, The training data adjusted by the output signal of the blood pressure neural network 66 is a pure value and has no quality. Traditional algorithmic processing considers both quality and artifact signals on the same basis. Major error sources in the process are overcome.

For those in the same field, the various examples disclosed are representative examples and would limit the scope of the claim. I think you can see that it's not a thing.

Engraving (no changes to the content) (Conventional technology) Fig, 5 Cuff pressure (mmHg) Fig, 7 Fig-8 Fig, 9 Systolic blood pressure (mmHg) Number of adaptive training buses Number of hidden nodes Number of solar calendar nodes a clean signal 10 noise signal Fig, 15 (one-way) 1.Display of the incident Blood pressure determination method and device University of Utah 4-Agent 6--Number of claims increased due to Ml addition 7-Object of amendment translation of the drawing international search report Continuation of front page (81) Designated countries EP (AT, BE, CH, DE.

DK, ES, FR, GB, GR, IT, LU, NL, SE), 0A (BF, BJ, CF, CG, CI, CM, GA, GN, ML, MR, SN, TD , TG), AT, AU, BB, BG, BR, CA, CH, C3, DE, DK.

ES, FI, GB, HU, JP, KP, KR, LK, LU, MC, MG, MN, MW, NL, No, PL, RO, SD, SE, SU □ (72) Inventor Oher, Joseph A.

University, Salt Lake City, Utah 84108, United States - Pileggi 1219 (72) Inventor: Westenskow, Dwayne Earl.

Winesob, Salt Lake City, Utah, 84121, USA Lord 172) Inventor Nygebert, Timothy P.

Salt Lake City, Utah, USA 84108 - Nis, Futhi Le Drive 1182

Claims (1)

[Claims] 1. For indirect quantitative assessment of variable physiological parameters without direct measurements. (1.1) identify a physiological parameter to be quantitatively monitored and evaluated; (1.3) a computer system that generates a signal sequence unsuitable for giving a reading; the neural network is capable of generating at least one output signal of the waiting input signal, the output signal providing an estimated value of the physiological parameter corresponding to the input signal; .4) determine the actual true value of the physiological parameter simultaneously with step 1.2, and (1.5) make adjustments within the neural network to adjust the physiological parameter determined in step 1.4. (1.6) (i) input signal, (ii) output signal to match the true parameter value. (iii) record the corresponding true values associated with the signals generated in step 1.2 in the memory of the computer system as training data; (1.7) the neural network; Processes input signals related to input signals and computer notes to evaluate physiological parameter values based on their association with training output signals within the A method for indirect, quantitative evaluation of variable physiological parameters, comprising the steps of sequentially repeating steps 1.2 to 1.6 as many times as necessary. 2. 2. The method of claim 1, wherein steps 1.5 and 1.6 include the detailed steps: (2.1) applying weighting factors to the interconnection nodes in the neural network received at the input nodes; (2.2) storing in computer memory the mutually coupled nodal weighting factors; An indirect quantitative evaluation method for variable physiological parameters consisting of: 3. A method according to claim 1, wherein step 2.1 comprises the detailed step of generating a signal sequence that is oscillatory but changes in amplitude or frequency in accordance with changes in the physiological parameter. 4. 4. The method of claim 3, further comprising: (4.1) selecting a plurality of sample signals from the signal sequence to be processed by the neural network; and (4.2) selecting at least one of the sample signals that can be processed by the neural network. Characteristics of A variable physiological parameter, possibly consisting of an additional step, is identified as a symptom signal. Indirect quantitative evaluation method for data. 5. 5. The method of claim 4, wherein step 4.2 includes the step of identifying the amplitude of the vibration signal as a feature defining a characteristic signal, and the method further comprises developing a waveform based on the sequential signals generated in step 1.2. wherein the waveform represents a locus of points representing the amplitude of each vibrational signal with respect to cuff pressure over a period of time consisting of a single diagnostic test procedure. 6. 6. The method of claim 5, wherein step 4.2 comprises performing a step for a single test procedure. Select at least 2 but not more than 10 sample signals from all the signals generated according to step 1.2, and apply the amplitude characteristic signals of these sample signals to the remaining steps of claim 1. Therefore, the entire signal sequence generated by processing is unsampled to produce the desired output signal. Variable physiology consists of detailed steps in training neural networks to identify signals. Indirect quantitative evaluation method of scientific parameters. 7. 6. The method of claim 5, wherein steps 1.2 and 1.3 include the following detailed steps: (7.1) develop a waveform for each single diagnostic test procedure consisting of a predetermined number of sample signals that approximately corresponds to the number of input nodes present in the neural network; (7.2) (7.3) Sample signal of the stored waveform Variable physiology consists of simultaneously and collectively transmitting signals to each input node of a neural network. Indirect quantitative evaluation method of scientific parameters. 8. 8. The method of claim 7, wherein step 4.1 comprises performing a step for a single test procedure. The whole of the signal sequence generated by selecting at least two sample signals from the total signals generated according to step 1.2 and processing the amplitude characteristic signals of these sample signals according to the remaining steps of claim 1. A method for indirect quantitative evaluation of variable physiological parameters consisting of detailed steps of training a neural network to identify a desired output signal without 9. 2. The method of claim 1, comprising the detailed step of identifying a physiological parameter and monitoring and evaluating it as a blood pressure parameter selected from a group of diastolic, mean and systolic intra-arterial blood pressures. Evaluation method. 10. The blood pressure parameter evaluation method according to claim 9, further comprising: (10.1) detecting the heart rate from an external pressure sensing means connected to the patient's body near the heartbeat sensing position; (10.2) identifying pulse amplitude characteristics of the vibration measurement signal; and (10.3) generating a vibration measurement signal sequence representative of a beat; measuring and recording pressure values (10.4) and generating an oscillatory signal representative of a heartbeat giving an accurate blood pressure reading; (10.5) send the values from step 10.3 to the input nodes of the neural network; (10.6) make adjustments within the neural network and perform step 10.4. to blood pressure determined (10.7) repeating these steps and the additional steps of claim 1 to recognize the associated vibration measurement input signals and a variable comprising an additional step of training the neural circuitry to evaluate blood pressure based on stored training data and an accumulation of training data stored in the computer system correlating the signal input with the actual blood pressure value; Indirect quantitative evaluation method of physiological parameters. 11. 11. The method of claim 10, wherein step 10.3 further comprises measuring and recording pressure values in predetermined increments over a pressure range of approximately 20 to 200 torr. A method for indirect quantitative evaluation of variable physiological parameters consisting of 12. 12. The method of claim 11, wherein step 10.3 comprises over a range of pressure values. Detailed steps for measuring and recording pressure values in evenly spaced increments of approximately 4 torr. A method for indirect quantitative evaluation of variable physiological parameters. 13. 12. The method of claim 11, further comprising storing the recorded pressure values in a memory, simultaneously transmitting the stored values to input nodes of the neural network, and processing the inputs received at the input nodes to perform actual processing. A single output signal corresponding to the invasive blood pressure parameter value Indirect quantitative evaluation of variable physiological parameters consisting of detailed steps of generating signals. value method. 14. 14. The method of claim 13, wherein the signal generated in step 10.1 is Variable physiology including detailed steps to select and send some to the input nodes of the neural network Indirect quantitative evaluation method of scientific parameters. 15. 15. The method of claim 14, wherein at least two representative signals are selected from the signals generated in step 1.2, and the total signal is divided into all interactive nodes of the neural network. Details of evaluating the waveform developed by the signal sent to the input nodes of the neural network without processing it at the point and generating the desired output signal corresponding to the blood pressure parameter. Quantitative evaluation method. 16. Online indirect quantitative assessment of variable physiological parameters without direct measurement (16.1) the physiological parameter to be quantitatively monitored and evaluated; identify the meter, (16.2) generate an on-line signal sequence that is quantitatively dependent on the physiological parameter but is not suitable for giving a direct quantitative reading based on direct measurement of the parameter, and (16.3) characterize the transmitting the signal containing the signal to the computer system containing the input node of the computer system assisted neural network, the neural network responding to the combined input signal; can generate a single output signal that is connected to the online signal. (16.4) processing the on-line signals in the neural network with respect to the training data to determine the physiological parameters associated with the signals; A method for online indirect quantitative evaluation of variable physiological parameters, comprising the steps of: identifying an evaluation value of a physiological parameter. 17. 2. The method of claim 1, wherein step 16.4 includes interconnections within the neural network. Pre-stored mutual connection weight coefficients as part of the data training to be applied to the joint points Details of applying the numbers to establish inter-nodal relationships as part of processing step 16.4. An indirect quantitative evaluation method for variable physiological parameters consisting of detailed steps. 18. 18. The method of claim 17, further comprising (18.1) a signal sequence or (18.2) selecting a plurality of sample signals from the neural network and processing such sample signals with a neural network trained to associate the relevant values of the physiological parameters; between variable physiological parameters consisting of the step of identifying at least one feature in the characteristic signal as a feature signal. Contact amount evaluation method. 19. 19. The method of claim 18, wherein step 18.2 comprises determining the amplitude of the vibration signal. identifying the characteristic signal as a defining feature; developing a waveform based on the sequential signals generated in step 16.2, said waveform being represented by a locus of points representing the amplitude of each vibration signal with respect to cuff pressure over a period of time consisting of a single diagnostic test procedure. variable physiological parameters An indirect quantitative measurement method for data. 20. 19. The method of claim 18, wherein step 18.2 is for a single test procedure. at least two sample signals from the total signal generated according to step 16.2. the amplitude characteristic signals of these sample signals in accordance with the remaining steps of claim 16 to evaluate the parameter values without processing all of the generated signals. Indirect quantitative measurement method. 21. 20. The method of claim 19, wherein steps 16.2 and 16.3 include the following detailed steps: (21.1) approximately corresponding to the number of input nodes in the neural network. (21.2) storing the sample signals in memory; and (21.3) applying the stored sample signals of the waveform to the neural circuit. A method for indirect quantitative evaluation of variable physiological samples, comprising: transmitting to each input node of a network. 22. 22. The method of claim 21, wherein step 21.2 relates to a single test procedure. at least two sample signals from the total signal generated according to step 16.2. and processing the amplitude and feature signals of these sampled signals according to the remaining steps of claim 16 to provide on-line signals to a neural network and generate a generated signal sequence. A method for indirect quantitative evaluation of variable physiological parameters, consisting of detailed steps of identifying the parameter being evaluated by simply sampling a few representative signals from the 23. 17. The method of claim 16, wherein diastolic, mean and systolic intra-arterial blood pressure Physiological parameters to be monitored and evaluated as blood pressure parameters selected from the group consisting of: Indirect quantitative assessment of variable physiological parameters consisting of detailed steps of identifying value method. 24. 24. The method of evaluating blood pressure parameters according to claim 23, further comprising: (24.1) generating a sequence of vibration measurement signals representative of heartbeats from an external pressure sensing means connected to the patient's body proximate to the heartbeat sensing location; , (24.2) identify the feature resulting in a pulse amplitude in the vibration measurement signal; (24.3) measure and record the pressure value in the pressure sensing means with the corresponding pulse amplitude signal of step 24.2; 24.4) Send pressure values and corresponding pulse amplitude signals to the input nodes of the computer system and its supported neural network, and create a database of online sample signals and training signals stored in the computer memory. (24.5) process the online sample signal in the neural network to determine the physiological parameter associated with the sample signal; A method for indirect quantitative evaluation of variable physiological parameters, comprising: determining the evaluation value of the data. 25. 25. The method of claim 24, wherein step 24.3 comprises a detailed step of measuring and recording pressure values in predetermined increments over a pressure range of approximately 20 to 200 torr. A method for indirect quantitative evaluation of variable physiological parameters. 26. 25. A method according to claim 24, wherein step 24.3 comprises applying pressure over the entire range of pressure values. A specific step of measuring and recording pressure values in evenly spaced increments of approximately 4 torr. A method for indirect quantitative measurement of variable physiological parameters. 27. 27. The method of claim 26, further comprising storing the recorded pressure values in memory. simultaneously transmit the recorded pressure values to the input nodes of the neural network, and receive the recorded pressure values at the input nodes. processing the received input to produce a single output signal corresponding to the estimated blood pressure parameter value; A method for indirect quantitative evaluation of variable physiological parameters consisting of a specific step of determining the 28. 28. A method according to claim 27, comprising the detailed step of selecting and transmitting some of the generated signals of step 24.1 to the input nodes of the neural network. Indirect quantitative evaluation method for physical parameters. 29. 29. The method of claim 28, wherein representative samples of at least two generated signals select a ring and process all signals at all interactive nodes of the neural network. The detailed steps of evaluating the waveforms developed by the signals sent to the input nodes of the neural network and generating the desired output signals corresponding to the blood pressure parameters without A method for indirect quantitative evaluation of variable physiological parameters. 30. 2. The method of claim 1, further comprising pre-preparing the generated signal received at the input node. The neural networks can be trained to recognize different physiological classifications of patients based on distinguishable patterns in the signals they receive, and are available online. Future comparisons with training data generated by patients falling into recognized patient categories as in-signals Indirect quantitative evaluation of variable physiological parameters consisting of an additional step of range narrowing value method. 31. 31. The method of claim 30, wherein the additional step includes pre-classifying the generated signal as representative of an adult or child patient. 32. 2. The method of claim 1, further comprising pre-preparing the generated signal received at the input node. Similarly, training neural networks to recognize faults as indicative of a fault occurring in diagnostic equipment connected to the patient and reject them as incorrect values of physiological parameters is an adjunct. An indirect quantitative evaluation method for variable physiological parameters consisting of an addition step. 33. 2. The method of claim 1, further comprising pre-preparing the generated signal received at the input node. Similarly, it is recognized as representing the occurrence of a signal modified by external noise, and physiological neural circuits to reject such altered signals as inaccurate values of biological parameters. A method for indirect quantitative evaluation of variable physiological parameters consisting of the additional step of training a network. 34. 34. The method of claim 33, wherein training the neural network to recognize the modified signal comprises the steps of: (34.1) generating an additional signal known to be modified; .2) sending the modified signal to the input node of the neural network; (34.3) applying a weighting factor to the interconnected nodes in the neural network and receiving the variable signal and the modified signal at the input node; (34.4) establish a nodal surface relationship between the desired output signal having a value defined to represent the interconnected nodal load factor in conjunction with the output value defined for the modified signal. (34.5) Repeat steps 34.1-34.4 sequentially to recognize similar modified signals. This includes the step of developing a statistical sample of the training data to A method for indirect quantitative evaluation of variable physiological parameters. 35. 35. The method of claim 34, wherein the patient follows step 1.2 of claim 1. A method for indirect quantitative evaluation of a variable physiological parameter, in which a modified signal is generated by inducing a modified signal within a signal generated by the method. 36. 35. The method of claim 34, wherein the steps are applied to a second neural network independently of the first neural network used to evaluate the physiological parameter value, the second neural network being the first neural network. A method for indirect quantitative evaluation of variable physiological parameters that serves to determine whether the values emanating from the output nodes of the neural circuitry of the brain should be saved or discarded as artifacts. 37. stored in the memory and supporting computer system of a neural network and useful for evaluating approximations of physiological parameters and according to the steps of claim 1; Training data generated by 38. In the training data according to claim 37, a method for non-invasively determining mean blood pressure. Training data with weighting factors applied to interconnected nodes in the neural network as part of the method. 39. 38. The training data of claim 37, comprising training data having weighting factors applied to interconnection nodes within a neural network as part of a method for non-invasively determining systolic blood pressure. Training data. 40. 38. The training data of claim 37, comprising training data having weighting factors applied to mutual waiting nodes within a neural network as part of a method for non-invasively determining diastolic blood pressure. Training data. 41. 35. The training data of claim 34, comprising weighting factors applied to interconnection nodes within the neural network as part of the method for determining the presence of modified signals to be discarded. training data. 42. In a device that indirectly monitors or evaluates a variable physiological parameter without the need for direct invasive measurements, the device (42.1) indirectly detects changes in the physiological parameter to be quantitatively monitored or evaluated. (42.2) a sensing means connected to the sensing means and fixed to a variable physiological parameter; Although quantitatively dependent, it is suitable for making direct quantitative readings based on direct measurements of parameters. (42.3) a neural network and a computer system connected to the generating means; Thus, the neural network comprises: (i) a series of input nodes receiving signals from the generating means; (ii) a series of hidden nodes individually connected to each input node; and (iii) each hidden node. including at least one output node connected thereto, the neural network including means for generating a single output signal from a signal received at the input node, the output signal being a physiological parameter. (42.4) an input signal received at the input node and a physiology specified during such training; (42.5) data storage means for storing training data generated within the neural network relating to a relationship with a desired value of a mathematical parameter; 1. A device for indirectly monitoring and evaluating a variable physiological parameter, comprising: reading means for indicating an evaluated value of the physiological parameter. 43. The apparatus according to claim 42, further comprising (43.1) a computer system. The physiological parameter is connected to the system and simultaneously generates the sample signal from the generation means. (43.2) invasive detection means adapted to the means for determining the actual true value of the parameter; an indirect monitoring and evaluation device for variable physiological parameters, comprising means for storing in a system and recalling and associating when a similar set of input signals recurs at input nodes of a neural network in the future. 44. 46. The apparatus of claim 45, wherein the neural network comprises at least two input nodes forming a hidden layer between the input nodes and a single output point. Direct monitoring and evaluation equipment. 46. 43. The apparatus of claim 42, further comprising a plurality of samples from the signal sequence. A selection control means for selecting a signal and processing it by a neural network is implemented in a computer system. The selection control means has a small number of signals that can be processed by neural circuits as characteristic signals. An apparatus for indirect monitoring and evaluation of a variable physiological parameter, comprising feature identification means for identifying at least one feature in each sample signal, said feature being associated with a changed value of the physiological parameter. 47. 43. The apparatus of claim 42, wherein the neural network measures mean arterial blood pressure, systolic blood pressure, Precepts stored within the computer system enable the neural network to determine the value of a physiological parameter selected from the group consisting of intrapulmonary blood pressure and diastolic arterial blood pressure. An indirect monitoring and evaluation device for variable physiological parameters, including performance data. 48. 43. The device of claim 42, wherein the device comprises three neural networks each configured and trained to determine mean intra-arterial blood pressure, systolic intra-arterial blood pressure and diastolic intra-arterial blood pressure. indirect monitoring and evaluation device for physical parameters. 49. 49. The apparatus of claim 48, further comprising a general input of an input node of the neural network. A set of generated signals representing force data is assigned to a classification identified by a preclassification neural network. an input signal suitable for processing within a second neural network trained with corresponding input data; an indirect monitoring and evaluation device for variable physiological parameters, including additional preclassification neural circuitry configured and trained to preclassify the variable as corresponding to a number.