Multi-sensor pain measurement system and evaluation method thereof

Abstract

The invention discloses a multi-sensor pain measurement system and an evaluation method thereof. The evaluation method comprises the following steps: preprocessing physiological characteristic parameters of a human body surface obtained in advance; and processing the preprocessed physiological feature parameters of the human body surface by using a physiological feature vector clustering method, a physiological signal feature vector regression model method or a physiological signal feature vector nonlinear regression method, and outputting the pain level of the patient. According to the invention, the mapping relation model of the physiological signal and the pain magnitude based on the artificial intelligence model is provided, and the multi-modal information with various forms, different sampling rates and different data volumes is converted into the pain magnitude which can be continuously valued in a real number field; therefore, an accurate and reliable pain measurement scheme is provided for dose-effect relationship evaluation of clinical application and analgesia treatment.

Description

Technical Field

[0001] The present invention relates to a multi-sensor pain measurement system and an evaluation method thereof, and belongs to the technical fields of pain medicine, human physiological function perception, multimodal information processing and artificial intelligence. Background Art

[0002] According to the medical definition, pain is an unpleasant sensory and emotional experience associated with actual or potential tissue damage. Pain is generally always accompanied by pain when the human body is sick, such as various inflammations.

[0003] Pain is a subjective experience that is affected by many factors, including physiology, psychology, personal experience, and social culture. In addition, individuals have different understandings and perceptions of pain. Currently, pain assessment in clinical practice is mainly based on self-rating scales, behavioral tests, and physiological measurements. Among them, the visual analogue scale (VAS) is the most commonly used single-dimensional measurement and assessment tool for pain intensity. The scale is mainly composed of a 100mm straight line, one end of which represents "completely painless" and the other end represents "the most intense pain imaginable" or "extreme pain". Patients will be asked to mark the corresponding position on this line to represent the intensity of pain they feel at the time. However, VAS requires patients to have a certain degree of abstract thinking ability, and the scoring is very subjective.

[0004] Correctly and objectively assessing pain is critical to the diagnosis of the patient's disease and the formulation and implementation of subsequent treatment plans. In addition, for studying the dose-effect relationship of a certain pain treatment plan, objective and quantitative pain assessment indicators are even more important because they determine the accuracy and credibility of the treatment effect evaluation. Obviously, the current self-assessment scale method is mainly based on subjective evaluation, scoring, and inquiry, which is not objective and quantitative, bringing uncertainty to diagnosis and subsequent treatment, and is more difficult to use for rigorous dose-effect relationship modeling and analysis.

[0005] At present, the methods used to measure and evaluate pain are mainly based on measuring physiological parameters of the human body. Common physiological parameters include electrodermal (EDA), electromyographic (EMG), electrocardiogram (ECG), electroencephalogram (EEG), heart rate (HR), pulse wave (PPG), and visual facial expression recognition. However, these signals have their own advantages and disadvantages, and it is difficult to ensure the accuracy of pain measurement when used alone. For example, EDA electrodermal is highly responsive to changes in sympathetic autonomic nerves, and is small in size, unobstructed, and highly mobile; but it cannot distinguish between pressure and pain, and cannot distinguish the source of pain (for example, the left or right side of the body). ECG electrocardiogram is sensitive to sympathetic and parasympathetic nerve responses, and its time domain and frequency domain characteristics are effective for evaluating acute pain, but it can be affected by positive or negative emotions (such as surprise, fear, or stress).

[0006] The existing technology provides a pain level assessment method and system, which establishes the brain network corresponding to each sample data through the sample data of the pain sample, the control sample and the induced sample, and calculates the topological attribute value of each brain network model, and obtains the dynamic variation gradient attribute of the brain network corresponding to the pain sample data relative to the brain network corresponding to the control sample data and the weight value of each brain region participating in pain perception, thereby obtaining a dynamic pain marker, and assessing the degree of pain perception of the brain of chronic pain patients based on the dynamic pain marker. Obviously, although this technology may establish a mapping relationship between the dynamic variation gradient attribute of the brain network and the participation of each brain region in pain perception, it is obviously very difficult to obtain a large number of brain network parameters corresponding to each brain region, and it requires the use of complex and expensive measurement equipment, which is not suitable for the measurement and evaluation of the general pain treatment effect.

[0007] The existing technology also provides a multi-dimensional measurement pain sensitivity assessment system. The control module integrates the control electrical stimulation module and the audio acquisition module. The electrical stimulation module gives the user pain stimulation according to parameters such as current intensity and pulse width. The acquisition module collects EEG signals and audio signals. The processing module processes the collected EEG signals and audio signals to obtain processing results. The stimulus-response model module incorporates the obtained results into the analysis and establishes a personalized stimulus-response model. The pain assessment integration module integrates the questionnaire assessment and the stimulus-response model data to calculate the individual pain sensitivity score. Obviously, this is a method of generating pain through electrical stimulation and then assessing pain through EEG signals. It is not based on the need to assess pain when diagnosing or treating the patient's pain, and the assessment method is still limited to EEG as a physiological signal.

[0008] The existing technology provides a method for estimating the magnitude of pain based on brain waves, which performs multiple brain wave measurements to obtain multiple brain wave data or analysis data thereof; based on the linearity of the relationship between the brain wave data or analysis data thereof (e.g., amplitude) and pain, the relative magnitude of pain when multiple brain wave measurements are performed is estimated based on the multiple brain wave data or analysis data thereof (e.g., amplitude). Obviously, this method also performs pain assessment based on a single signal, namely, brain waves.

[0009] The existing technology provides a method for measuring pain, which determines the reference stimulation intensity corresponding to the reference pain of the subject based on the pain level reported by the subject and the stimulation intensity applied to the subject. Obviously, this is also a method of generating pain by applying stimulation and then evaluating pain through EEG signals. It is not based on the need to evaluate pain when diagnosing or treating the patient's pain, and the evaluation method is still limited to EEG, a physiological signal.

[0010] As can be seen from the above technologies, existing pain measurement technologies mainly evaluate pain based on a single physiological signal, or use a complex and costly EEG measurement network to achieve a high degree of credibility. Therefore, it is impossible to achieve accurate and reliable pain measurement and evaluation at a low cost. At the same time, measuring EEG signals requires the use of a relatively bulky head-mounted electrode network, which is not easy to implement in clinical or home applications. Summary of the invention

[0011] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a multi-sensor pain measurement system and an evaluation method thereof, so that different sensors can complement each other's advantages, without the need to obtain a large number of brain network parameters corresponding to each brain area, without the need to use complex and expensive measurement equipment, and not limited to a physiological signal of electroencephalogram. Pain assessment is not based on a single signal of electroencephalogram, but rather uses parameters of at least four sensors to assess pain; at the same time, sensors with lower cost and smaller size, such as HR, EDG, EMG, and ECG, are preferably used to realize a portable and wearable pain measurement device; a corresponding mapping relationship model between physiological signals and pain magnitudes based on an artificial intelligence model is proposed, and multimodal information with various forms, different sampling rates, and different data volumes is converted into a pain magnitude that can be continuously valued in a real number domain, thereby providing an accurate and reliable pain measurement scheme for clinical applications and dose-effect relationship evaluation of analgesic treatment.

[0012] Preferably, the present invention provides a multi-sensor pain assessment method comprising:

[0013] Preprocessing the physiological characteristic parameters of the human body surface obtained in advance;

[0014] The preprocessed physiological characteristic parameters of the human body surface are processed using a physiological characteristic vector clustering method, a physiological signal characteristic vector regression model method or a physiological signal characteristic vector nonlinear regression method to output the patient's pain level.

[0015] Preferably, the pre-acquired physiological characteristic parameters of the human body surface are pre-processed, including:

[0016] Perform signal processing and time-frequency domain analysis on the physiological characteristic parameters of the historical human body surface to extract the time-frequency domain features;

[0017] All time-frequency domain features are taken as a set of historical feature vectors.

[0018] Preferably, the physiological characteristic parameters of the human body surface after preprocessing are processed using the physiological characteristic vector clustering method to output the patient's pain level, including:

[0019] The K-means clustering method is used to process the historical feature vector set to obtain the central points p1, p2, ..., p of the historical feature vector set corresponding to the k levels in the preset pain assessment scale.k .

[0020] Preferably, the physiological characteristic parameters of the human body surface after preprocessing are processed using the physiological characteristic vector clustering method to output the patient's pain level, including:

[0021] Acquire the physiological characteristic parameters of new patients and perform normalization processing;

[0022] Extract the time-frequency domain features of the normalized physiological characteristic parameters of the new patient to obtain a multidimensional feature vector q;

[0023] Calculate the eigenvector q and the center point set P = {p1, p2, ..., p k The distance d between each center point in i =||qp i ||2;

[0024] Calculate the pain level t of a new patient:

[0025] t=Σ i＝1 k T i ·(d i / Σ i＝1 k d i ),

[0026] Among them, Σ i＝1 k Sum from i=1 to i=k.

[0027] Preferably, the physiological signal feature vector regression model method is used to process the preprocessed physiological characteristic parameters of the human body surface and output the patient's pain level, including:

[0028] Extract the feature vectors of all patients in the historical feature vector set and the corresponding pain level The feature vector q of the i-th patient i =[q i1 ,q i2 ,…,q im ], i = 1, 2, ..., N, the pain level of the i-th patient is T i , the eigenvector q i is an m-dimensional row vector;

[0029] The parameter vector A is obtained by calculating the least squares solution:

[0030]

[0031] Among them, the superscript + sign indicates the pseudo-inverse of the matrix, N is the total number of patients in the historical feature vector set, and the total number of patients N is greater than the dimension m of the feature vector;

[0032] Normalizing the acquired physiological characteristic parameters of the new patient;

[0033] Extract the normalized time-frequency domain features of the physiological characteristic parameters of the new patient to obtain a multidimensional feature vector q;

[0034] Calculate the patient's pain level:

[0035] T = q·A.

[0036] Preferably, the pre-processed physiological characteristic parameters of the human body surface are processed using a physiological signal characteristic vector nonlinear regression method to output the patient's pain level, including:

[0037] Extract the feature vectors of all patients in the historical feature vector set and the corresponding pain level The feature vector q of the i-th patient i =[q i1 ,q i2 ,…,q im ], i = 1, 2, ..., N, the pain level of the i-th patient is T i , the eigenvector q i is an m-dimensional row vector; using n nonlinear basis functions f1,…,f n Perform nonlinear mapping on each eigenvector respectively, obtain n nonlinear basis function values ​​of each eigenvector of each patient, and calculate the relationship between the nonlinear basis function value and the pain level:

[0038]

[0039] Among them, B is the unknown vector, r Nn is the nth nonlinear basis function value of each eigenvector of the Nth patient. The superscript + sign indicates the pseudo-inverse of the matrix. The total number of patients N is greater than the dimension n of the nonlinear basis function.

[0040] Find the least squares solution of the above equation and get the parameter vector B:

[0041]

[0042] Obtain the physiological characteristic parameters of new patients and perform normalization processing;

[0043] Extract various features of the physiological characteristic parameters of the new patient to obtain a multi-dimensional feature vector q;

[0044] Calculate the patient's pain level T:

[0045] T=[r1,…,r n ]·B,

[0046] r j =f j (q),

[0047] Among them, f j (q) is the j-th nonlinear basis function value, j=1,…n.

[0048] Prioritize the signal processing and time-frequency domain analysis of the physiological characteristic parameters of the historical human body surface to extract and obtain the time-frequency domain features, including:

[0049] Filter and normalize the physiological characteristic parameters of the historical human body surface;

[0050] The time-frequency domain features used to assess pain are calculated using the normalized historical physiological characteristic parameters of the human body surface:

[0051] The time-frequency domain features include the statistical features and frequency domain features of the time domain signals. The time domain analysis method is used to extract the time domain features of the EDA signal, the time domain features of the EMG signal, the time domain features of the ECG signal and the time domain features of the HR signal. The statistical features of the time domain signals include SDNN, RMSSD, SDSD, NN50 and PNN50. The frequency domain analysis method is used to extract the frequency domain features of the EDA signal, the frequency domain features of the EMG signal, the frequency domain features of the ECG signal and the frequency domain features of the HR signal. The frequency domain features include total power, low-frequency power and high-frequency power.

[0052] Preferably, the statistical features of the time domain signal include SDNN, RMSSD, SDSD, NN50 and PNN50, and the frequency domain features include total power, low frequency power and high frequency power.

[0053] Preferably, physiological characteristic parameters of the human body surface are obtained in advance, including:

[0054] Physiological characteristic parameters of the human body surface including EDA signals, EMG signals, ECG signals and HR signals are acquired using skin electrode sensors, myoelectric sensors, electrocardiographic sensors and heartbeat sensors.

[0055] Preferably, the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the methods described in the first aspect when executing the program.

[0056] Preferably, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods described in the first aspect.

[0057] The beneficial effects achieved by the present invention are:

[0058] The present invention proposes a measurement system based on a variety of different sensors, and takes a physiological characteristic parameter signal characteristic vector composed of two eigenvalues ​​and a cluster corresponding to five pain magnitudes as examples to illustrate a method for establishing a mathematical model of a mapping relationship between physiological characteristic parameters and pain indicators based on clustering of physiological characteristic parameter characteristic vectors; different sensors can complement each other's advantages; at the same time, sensors with low cost and small size are preferably used, such as skin electrodeposition sensors, myoelectric sensors, electrocardiogram sensors and heartbeat sensors, to realize a portable and wearable pain measurement device; a corresponding mapping relationship model between physiological signals and pain magnitudes based on an artificial intelligence model is proposed, and multimodal information with various forms, different sampling rates and different data volumes is converted into a pain magnitude that can be continuously valued in a real number domain; thereby providing an accurate and reliable pain measurement scheme for clinical applications and dose-effect relationship evaluation of analgesic treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0060] Figure 1 is a clustering schematic diagram of physiological characteristic parameter signal feature vectors in some embodiments of the present application;

[0061] Figure 2 is a principle block diagram of physiological characteristic parameter acquisition and feature extraction in some embodiments of the present application;

[0062] Figure 3 is a schematic diagram of an ECG signal sampled in some embodiments of the present application;

[0063] Figure 4 It is a schematic diagram of an ECG signal after low-pass filtering in some embodiments of the present application. DETAILED DESCRIPTION

[0064] Existing pain measurement technologies mainly evaluate pain based on a single physiological signal, or use a complex and expensive brain wave measurement network to achieve a high degree of credibility. Therefore, it is impossible to achieve accurate and reliable pain measurement and evaluation at a low cost. At the same time, measuring EEG signals requires the use of a relatively bulky head-mounted electrode network, which is not easy to implement in clinical or home applications. Figure 1 , the present application discloses a multi-sensor pain assessment method, comprising:

[0065] Based on the physiological characteristics of pain symptoms on the human body surface, the physiological characteristic parameters of the human body surface are collected, and signal processing (normalization) and time-frequency domain analysis are performed to extract the time-frequency domain features as a set of historical feature vectors to evaluate pain.

[0066] A mathematical model of the mapping relationship between physiological characteristic parameters and pain indicators is established, and quantitative indicators that can be measured in real time are established.

[0067] Based on the physiological characteristics of pain symptoms on the human body surface, physiological characteristic parameters of the human body surface are collected, and signal processing and time-frequency domain analysis are performed to extract time-frequency domain features, including:

[0068] Methods and system principles Figure 1 As shown. As a specific embodiment, the system includes skin electricity sensor, myoelectric sensor, electrocardiogram sensor and heartbeat sensor, but the present technology is not limited to the use of these four sensors. The purpose of using these four sensors is to overcome the shortcomings of a single sensor. For example, ECG electrocardiogram can make up for the weakness of EDA skin electricity in responding to acute pain; EDA skin electricity can make up for the problem that ECG electrocardiogram is easily affected by positive or negative emotions.

[0069] The following uses the signal collected by the ECG sensor as an example to further illustrate the time-frequency domain analysis and feature extraction method. The same signal processing and statistical analysis methods are also used for the analysis and feature extraction of other sensors (skin electrical sensors, myoelectric sensors and heartbeat sensors).

[0070] Since the sampled original time domain signal contains high-frequency measurement noise, it is first filtered using a low-pass filter, such as a digital low-pass 5th-order Butterworth filter or a simple sliding average filter, to obtain a filtered ECG signal.

[0071] From the filtered ECG signal, the time-frequency domain features for pain assessment can be further calculated as follows:

[0072] The time-frequency domain features include the statistical features and frequency-domain features of the time-domain signal;

[0073] The statistical features of time domain signals include SDNN, RMSSD, SDSD, NN50, and PNN50. The calculation of these statistical features requires the use of ECG signals of multiple heart cycles:

[0074] SDNN: standard deviation of all normal sinus intervals (NN), unit: ms.

[0075] RMSSD: The root mean square value of the difference between adjacent normal sinus beat intervals, unit: ms.

[0076] SDSD: The standard deviation of the differences between all adjacent normal sinus beat intervals, unit: ms.

[0077] NN50: The number of heartbeats with a difference of more than 50 ms between adjacent normal sinus beat intervals among all normal sinus beat intervals, unit:

[0078] PNN50: NN50 divided by the total number of normal sinus intervals, multiplied by 100, unit: %.

[0079] Frequency domain features include total power, low-frequency power, and high-frequency power. To calculate these frequency domain features, the time domain ECG signal must first be subjected to a fast Fourier transform (FFT) and the ECG signal of multiple heart cycles must be used:

[0080] Total power (TP): frequency band ≤ 0.4HZ;

[0081] Low frequency power (LF): frequency band 0.04~0.15HZ;

[0082] High frequency power (HF): frequency band 0.15~0.4HZ.

[0083] Based on the physiological characteristics of pain symptoms on the human body surface, the physiological characteristic parameters of the human body surface are collected, and signal processing and time-frequency domain analysis are performed to extract the time-frequency domain features. The specific steps are as follows:

[0084] 1) Collect the patient's pain assessment scale score level;

[0085] 2) Collecting the patient's multimodal sensor signal data set, including the time series of EDA signal, EMG signal, ECG signal and HR signal; Figure 3 As shown in the figure, the original time domain signal obtained by sampling contains high-frequency measurement noise, so it is first filtered with a low-pass filter to obtain the following Figure 4 The ECG time domain signal is shown after low-pass filtering. The low-pass filtering can use a digital low-pass 5th-order Butterworth filter or a simple sliding average filter.

[0086] 3) low-pass filtering the time series including EDA signals, EMG signals, ECG signals and HR signals to filter out measurement noise;

[0087] 4) normalizing the time series including EDA signals, EMG signals, ECG signals and HR signals, that is, dividing the value of each signal time series by its maximum value;

[0088] 5) Using the time domain analysis method to extract the time domain features of the EDA signal, the EMG signal, the ECG signal and the HR signal. The time domain feature vector set of the EDA signal, the EMG signal, the ECG signal and the HR signal includes the set of SDNN, RMSSD, SDSD, NN50 and PNN50 of the EDA signal, the EMG signal, the ECG signal and the HR signal;

[0089] 6) The frequency domain analysis method is used to extract the frequency domain features of the EDA signal, the frequency domain features of the EMG signal, the frequency domain features of the ECG signal and the frequency domain features of the HR signal. The frequency domain feature vector sets of the EDA signal, the EMG signal, the ECG signal and the HR signal include the sets of the total power, the low-frequency power and the high-frequency power of the EDA signal, the EMG signal, the ECG signal and the HR signal; the time domain feature vector set and the frequency domain feature vector set are combined into a historical feature vector set.

[0090] Standard time domain analysis, statistical analysis, and frequency domain analysis methods can be used here. Because different signals have different processing methods, the present invention does not specifically limit these traditional feature extraction methods, but focuses on proposing a mathematical modeling method for the mapping relationship between feature vectors and pain levels. The time-frequency domain features extracted in the above steps can constitute a multi-dimensional feature vector, which is used for clustering or classification in the feature space to distinguish different discrete levels of pain levels, and can also be used to train a regression neural network model.

[0091] Based on the clustering of physiological feature vectors, a mathematical model of the mapping relationship between physiological feature parameters and pain indicators is established, including:

[0092] The method based on physiological feature vector clustering includes:

[0093] 1) The K-means clustering method is used to process the historical feature vector set to obtain the center points of the feature vector set corresponding to the ten levels in the preset pain assessment scale, denoted as p1, p2, ..., p 10 , save the center point set P = {p1, p2, ..., p 10}, the preset pain assessment scale may be a VAS scale;

[0094] 2) Obtain the physiological characteristic parameters of the new patient and perform normalization processing;

[0095] 3) extracting the normalized time-frequency domain features of the physiological characteristic parameters of the new patient and obtaining a multidimensional feature vector, denoted as q;

[0096] 4) Calculate the distance between the feature vector q and each center point in the center point set P, denoted as d i =||qp i ||2, that is, vector q and the i-th center p i The second norm of the difference;

[0097] 5) Calculate the pain level of the new patient according to the following weighted average formula: t = Σ i＝1 10 T i ·(d i / Σ i＝1 10 di ), where Σ i＝1 10 The sum is calculated from i=1 to i=10, that is, the formula uses the ratio of the distance between the patient's physiological characteristic parameter feature parameter vector and the 10 pain level cluster centers as the weight, and calculates an average value as the patient's pain level. Since the weighted average is calculated, the calculated level must be a real number between 1 and 10.

[0098] Method based on physiological signal feature vector regression model:

[0099] 1) Extract the feature vectors of all patients in the historical feature vector set and the corresponding pain level The feature vector q of the i-th patient i =[q i1 ,q i2 ,…,q im ], i = 1, 2, ..., N, the pain level of the i-th patient is T i , the eigenvector q i is an m-dimensional row vector. Here, we assume that there are m eigenvalues, that is, the eigenvector is an m-dimensional row vector. The pain level corresponding to the eigenvector of the i-th patient is recorded as T i ; Suppose there is an unknown parameter vector A, so that the following equation holds:

[0100]

[0101] The superscript + sign indicates the pseudo-inverse of the matrix. Here, it is assumed that the number of patients N in the training data set is greater than the dimension m of the feature vector, and the rank of the matrix is ​​m;

[0102] 2) Find the least squares solution of the above equation and obtain the parameter vector A, that is:

[0103]

[0104] 3) Obtain the physiological characteristic parameters of the new patient and perform normalization processing;

[0105] 4) extracting the normalized time-frequency domain features of the physiological characteristic parameters of the new patient to form a multidimensional feature vector, denoted as q;

[0106] 5) Calculate T = q·A and find the patient's pain level T.

[0107] The above regression model-based method can also be extended to nonlinear regression to improve the complexity and potential accuracy of the model. The specific method is as follows:

[0108] 1) Extract the feature vector of the i-th patient in the historical feature vector set, denoted as q i =[q i1,q i2 ,…,q im ,] (here we assume that there are m eigenvalues, that is, the eigenvector is an m-dimensional row vector), and the corresponding pain level is recorded as T i ;

[0109] 2) Select n nonlinear basis functions (such as polynomial functions, radial basis functions, etc.) f1,…,f n , use n nonlinear basis functions to perform nonlinear mapping on each eigenvector and obtain the nonlinear basis function value: r ij =f j (q i ), so the eigenvector corresponding to the i-th patient generates n nonlinear basis function values, and the relationship between the N nonlinear basis function values ​​and the pain level is expressed as the following linear combination, where B is an unknown parameter column vector:

[0110]

[0111] The superscript + sign indicates the pseudo-inverse of the matrix. It is assumed here that the number of patients N in the data set is greater than the dimension n of the nonlinear basis function, and the rank of the matrix is ​​n;

[0112] 3) Find the least squares solution of the above equation and get the parameter vector B, that is:

[0113]

[0114] 4) Obtain the physiological characteristic parameters of the new patient and perform normalization processing;

[0115] 5) extracting various features of the physiological characteristic parameters of the new patient to obtain a multi-dimensional feature vector, denoted as q;

[0116] 6) For j = 1, ..., n, calculate the nonlinear basis function value r j =f j (q), calculate T = [r1,…,r n ]·B, calculate the patient's pain level T.

[0117] In an embodiment of the present application, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the above methods when executing the program.

[0118] In an embodiment of the present application, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0119] like Figure 1As shown, the method of establishing a mathematical model of the mapping relationship between physiological characteristic parameters and pain indicators based on the clustering of physiological characteristic parameter feature vectors is illustrated by taking the physiological characteristic parameter signal feature vector composed of two eigenvalues ​​and the clusters corresponding to five pain levels as examples. Clusters 1 to 5 represent pain levels 1 to 5, respectively. The physiological characteristic parameter vector of the newly measured patient is represented by a square, and its distance to the center points of the five level clusters is d1,…,d5, respectively. The patient's pain level t can be calculated according to the following weighted average formula:

[0120] t=Σ i＝1 5 T i ·(d i / Σ i＝1 5 d i ).

[0121] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0122] Those skilled in the art will readily appreciate other embodiments of the invention after considering the specification and practicing the invention invented herein. This application is intended to cover any variations, uses or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art that are not invented by the present invention. The specification and examples are intended to be exemplary only, and the true scope and spirit of the invention are indicated by the claims.

[0123] The above specific implementation methods further illustrate the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above are only specific implementation methods of the present application and are not used to limit the protection scope of the present application. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solutions of the present application should be included in the protection scope of the present application.

Claims

1. A multi-sensor pain assessment method, characterized in that: include: Preprocessing the physiological characteristic parameters of the human body surface obtained in advance; The preprocessed physiological characteristic parameters of the human body surface are processed using a physiological characteristic vector clustering method, a physiological signal characteristic vector regression model method or a physiological signal characteristic vector nonlinear regression method to output the patient's pain level.

2. A multi-sensor pain assessment method according to claim 1, characterized in that: Preprocessing the physiological characteristic parameters of the human body surface obtained in advance includes: Perform signal processing and time-frequency domain analysis on the physiological characteristic parameters of the historical human body surface to extract the time-frequency domain features; All time-frequency domain features are taken as a set of historical feature vectors.

3. A multi-sensor pain assessment method according to claim 1, characterized in that: The physiological characteristic parameters of the human body surface after preprocessing are processed using the physiological characteristic vector clustering method to output the patient's pain level, including: The K-means clustering method is used to process the historical feature vector set to obtain the central points p1, p2, ..., p of the historical feature vector set corresponding to the k levels in the preset pain assessment scale. k .

4. A multi-sensor pain assessment method according to claim 3, characterized in that: The physiological characteristic parameters of the human body surface after preprocessing are processed using the physiological characteristic vector clustering method to output the patient's pain level, including: Acquire the physiological characteristic parameters of new patients and perform normalization processing; Extract the time-frequency domain features of the normalized physiological characteristic parameters of the new patient to obtain a multidimensional feature vector q; Calculate the eigenvector q and the center point set P = {p1, p2, ..., p k The distance d between each center point in i =||qp i ||2; Calculate the pain level t of a new patient: t=Σ i＝1 k T i ·(d i / S i＝1 k d i ), Among them, Σ i＝1 k Sum from i=1 to i=k.

5. A multi-sensor pain assessment method according to claim 1, characterized in that: The physiological signal feature vector regression model method is used to process the pre-processed physiological characteristic parameters of the human body surface and output the patient's pain level, including: Extract the feature vectors of all patients in the historical feature vector set and the corresponding pain level The feature vector q of the i-th patient i =[q i1 ,q i2 ,…,q im ], i = 1, 2, ..., N, the pain level of the i-th patient is T i , the eigenvector q i is an m-dimensional row vector; The parameter vector A is obtained by calculating the least squares solution: Among them, the superscript + sign indicates the pseudo-inverse of the matrix, N is the total number of patients in the historical feature vector set, and the total number of patients N is greater than the dimension m of the feature vector; Normalizing the acquired physiological characteristic parameters of the new patient; Extract the normalized time-frequency domain features of the physiological characteristic parameters of the new patient to obtain a multidimensional feature vector q; Calculate the patient's pain level: T = q·A.

6. A multi-sensor pain assessment method according to claim 1, characterized in that: The physiological characteristic parameters of the human body surface after preprocessing are processed using the physiological signal feature vector nonlinear regression method to output the patient's pain level, including: Extract the feature vectors of all patients in the historical feature vector set and the corresponding pain level The feature vector q of the i-th patient i =[q i1 ,q i2 ,…,q im ], i = 1, 2, ..., N, the pain level of the i-th patient is T i , the eigenvector q i is an m-dimensional row vector; using n nonlinear basis functions f1,…,f n Perform nonlinear mapping on each eigenvector respectively, obtain n nonlinear basis function values ​​of each eigenvector of each patient, and calculate the relationship between the nonlinear basis function value and the pain level: Among them, B is the unknown vector, r Nn is the nth nonlinear basis function value of each eigenvector of the Nth patient. The superscript + sign indicates the pseudo-inverse of the matrix. The total number of patients N is greater than the dimension n of the nonlinear basis function. Find the least squares solution of the above equation and get the parameter vector B: Obtain the physiological characteristic parameters of new patients and perform normalization processing; Extract various features of the physiological characteristic parameters of the new patient to obtain a multi-dimensional feature vector q; Calculate the patient's pain level T: T=[r1,…,r n ]·B, r j =f j (q), Among them, f j (q) is the j-th nonlinear basis function value, j=1,…n.

7. A multi-sensor pain assessment method according to claim 3, characterized in that: Perform signal processing and time-frequency domain analysis on the physiological characteristic parameters of the historical human body surface to extract the time-frequency domain features, including: Filter and normalize the physiological characteristic parameters of the historical human body surface; The time-frequency domain features used to assess pain are calculated using the normalized historical physiological characteristic parameters of the human body surface: The time-frequency domain features include the statistical features and frequency domain features of the time domain signals. The time domain analysis method is used to extract the time domain features of the EDA signal, the time domain features of the EMG signal, the time domain features of the ECG signal and the time domain features of the HR signal. The statistical features of the time domain signals include SDNN, RMSSD, SDSD, NN50 and PNN50. The frequency domain analysis method is used to extract the frequency domain features of the EDA signal, the frequency domain features of the EMG signal, the frequency domain features of the ECG signal and the frequency domain features of the HR signal. The frequency domain features include total power, low-frequency power and high-frequency power.

8. A multi-sensor pain assessment method according to claim 3, characterized in that: The statistical features of time domain signals include SDNN, RMSSD, SDSD, NN50 and PNN50, and the frequency domain features include total power, low-frequency power and high-frequency power.

9. A multi-sensor pain assessment method according to claim 8, characterized in that: The physiological characteristic parameters of the human body surface are obtained in advance, including: Physiological characteristic parameters of the human body surface including EDA signals, EMG signals, ECG signals and HR signals are acquired using skin electrode sensors, myoelectric sensors, electrocardiographic sensors and heartbeat sensors.

Images