Chronic pain intensity intelligent evaluation system based on heart rate variability analysis

Abstract

The invention provides a chronic pain intensity intelligent assessment system based on heart rate variability analysis, comprising: a data acquisition module for acquiring electrocardiogram data of a patient in real time and extracting an HRV feature set in real time; the dimension reduction module is used for respectively carrying out manual screening and PCA dimension reduction on the extracted HRV feature set to obtain a first dimension reduction feature set and a second dimension reduction feature set; the feature splicing module is used for splicing and normalizing HRV features in the first dimension reduction feature set and the second dimension reduction feature set to obtain spliced features; the pain assessment module is used for inputting the splicing features into a pre-trained multi-level interactive attention network to assess the pain intensity level; according to the invention, more efficient, personalized and dynamic pain supervision service can be provided for patients and medical personnel in real time, the effect of pain management is improved, the overall medical experience is also improved, and a basis is provided for future medical research and technical development.

Description

Technical Field

[0001] The present invention relates to the technical field of medical devices, and more particularly, to an intelligent chronic pain intensity assessment system based on heart rate variability analysis. Background Art

[0002] Pain is an unpleasant sensory and emotional experience, usually caused by actual or potential tissue damage. According to the duration, pain can be divided into acute pain and chronic pain. Acute pain is an adaptive signal to prevent danger and ensure survival, while chronic pain is usually associated with chronic diseases and untreated medical pathologies. Currently, the most commonly used method for experimentally induced pain assessment is self-report measurement; for example, pain intensity can be measured by using a visual analogue scale (VAS), a numerical rating scale (NRS), and a verbal rating scale (VRS).

[0003] In recent years, the relationship between pain and heart rate variability (HRV) has received extensive attention. As a simple, easy-to-implement, and non-invasive means for quantitatively evaluating the function of the autonomic nervous system (ANS), with the continuous in-depth research, the application of clinical HRV is no longer limited to cardiovascular diseases, and it is also increasingly used in pain diseases. According to research, there is a complex interaction relationship between pain and cardiac electrical activity. Pain not only directly affects various indicators in the electrocardiogram, but also regulates cardiac activity through the autonomic nervous system. For example, pain usually triggers the activation of the sympathetic nervous system, resulting in an increase in heart rate. This is manifested in the electrocardiogram as a shortening of the heartbeat interval (i.e., shortening of the RR interval). In addition, pain usually leads to a decrease in heart rate variability, which is the result of increased sympathetic nerve activity and decreased vagus nerve activity. The decrease in HRV reflects the weakening of cardiac regulation function. And in some studies, it has been found that pain may affect the QT interval, which is an indicator representing the repolarization time of cardiac electrical activity in the ECG. The stress response caused by pain may lead to an extension or shortening of the QT interval. Therefore, electrocardiac activity can be used as an important tool for evaluating and managing pain.

[0004] The existing patent document with the publication number "CN119314703A" provides a real-time pain monitoring and feedback system, including: a pain biomarker monitoring module, an external environment monitoring module, a data transmission module, a data processing and analysis module, a feedback module, a medical professional support module, an intelligent reminder module, and a continuous learning and optimization module; although this existing technology performs real-time pain monitoring and feedback based on heart rate variability and skin electrical response, however, due to the large amount of electrocardiac data, and the existence of redundant features and noise, the processing speed of this existing technology is slow, the real-time performance is poor, and it cannot accurately reflect the true pain intensity level, and the calculation accuracy needs to be improved. Summary of the Invention

[0005] To overcome the defect that it is difficult to evaluate pain intensity in real time, efficiently, and accurately in the above-mentioned prior art, the present invention provides an intelligent chronic pain intensity evaluation system based on heart rate variability analysis. By combining statistical analysis methods, principal component analysis (PCA) dimensionality reduction technology, and multi-level interactive attention technology, it can calculate in real time the pain index represented by ECG data and realize quantitative evaluation of the pain degree of patients.

[0006] To solve the above technical problems, the technical solution of the present invention is as follows:

[0007] An intelligent chronic pain intensity evaluation system based on heart rate variability analysis, comprising:

[0008] A data acquisition module: used to collect the electrocardiogram data of patients in real time and extract the HRV feature set in real time, and the HRV feature set includes a number of HRV features;

[0009] A dimensionality reduction module: used to perform manual screening and PCA dimensionality reduction on the extracted HRV feature set respectively to obtain a first dimensionality reduction feature set and a second dimensionality reduction feature set;

[0010] A feature splicing module: used to splice and normalize the HRV features in the first dimensionality reduction feature set and the second dimensionality reduction feature set to obtain spliced features;

[0011] A pain evaluation module: used to input the spliced features into a pre-trained multi-level interactive attention network for evaluating the pain intensity level. The multi-level interactive attention network outputs in real time the pain index of the patient, realizing real-time intelligent evaluation of chronic pain intensity.

[0012] Preferably, in the data acquisition module, the electrocardiogram data of patients is collected in real time by the bipolar lead method of RA and LL positions;

[0013] The collected electrocardiogram data is denoised and R peak located to obtain the RR sequence data of the electrocardiogram data, and further calculate a number of features of HRV in the time domain, frequency domain, and non-linear domain, and extract the HRV feature set.

[0014] Preferably, the HRV features at least include: average heart rate interval, standard deviation of heart rate interval, standard deviation of differences between consecutive heart rate intervals, root mean square of differences between consecutive heart rate intervals, median of heart rate interval, logarithm of the number of differences between consecutive heart rate intervals greater than 50 ms, percentage of NN50 in the logarithm of the total heart rate interval, logarithm of the number of differences between consecutive heart rate intervals greater than 20 ms, percentage of NN20 in the logarithm of the total heart rate interval, difference between the maximum and minimum values of the heart rate interval, coefficient of variation of differences between consecutive heart rate intervals, coefficient of variation of the heart rate interval, average heart rate, maximum heart rate, minimum heart rate, standard deviation of heart rate, total power, very low frequency power with a frequency range of 0.0033 - 0.04 Hz, low frequency power with a frequency range of 0.04 - 0.15 Hz, high frequency power with a frequency range of 0.15 - 0.4 Hz, low frequency to high frequency power ratio, relative proportion of low frequency power in the total power, relative proportion of high frequency power in the total power, and regression slope of the RR sequence.

[0015] Preferably, in the dimensionality reduction module, several significant HRV features are manually selected from the HRV feature set to obtain the first dimensionality reduction feature set;

[0016] Using the PCA dimensionality reduction algorithm, all HRV features are projected onto the orthogonal coordinate system with the largest variance through linear transformation, and several principal component features are retained to obtain the second dimensionality reduction feature set.

[0017] Preferably, the first dimensionality reduction feature set includes the following 8 significant HRV features:

[0018] Standard deviation of heart rate interval, root mean square of differences between consecutive heart rate intervals, coefficient of variation of the heart rate interval, average heart rate, high frequency power with a frequency range of 0.15 - 0.4 Hz, low frequency to high frequency power ratio, relative proportion of high frequency power in the total power, and regression slope of the RR sequence.

[0019] Preferably, the PCA dimensionality reduction algorithm includes:

[0020] Perform standardization processing on all HRV features, calculate the covariance matrix of all standardized HRV features, perform eigenvalue decomposition on the covariance matrix to obtain several eigenvalues and their corresponding eigenvectors, where the eigenvalue represents the variance of each principal component, and the eigenvector represents the direction of the principal component;

[0021] Arrange the eigenvalues in descending order, and the corresponding eigenvectors are also arranged in the same order; the eigenvector with a larger eigenvalue represents a more important principal component;

[0022] Select the first 8 eigenvectors according to the size of the eigenvalues and construct a projection matrix;

[0023] Project all HRV features into the principal component space using the projection matrix to obtain the dimension-reduced HRV features, and jointly save the dimension-reduced HRV features as the second dimension-reduced feature set.

[0024] Preferably, in the pain assessment module, the spliced features are used as first-order features, and the outer product calculation is performed on the first-order features to obtain second-order features. Each element in the second-order features represents the relationship between sample points;

[0025] Input the first-order features and the second-order features into the pre-trained multi-level interactive attention network. A Transformer encoder and an interactive attention fusion unit are set in the multi-level interactive attention network;

[0026] The first-order features and the second-order features are first calculated by two groups of independent Transformer encoders to obtain the feature codes and classification codes of each order; subsequently, the obtained feature codes and classification codes of each order are jointly input into the interactive attention fusion unit for multi-level interactive fusion; in the interactive attention fusion unit, the classification codes of each order are respectively subjected to mutual attention calculation with the feature codes of different orders to obtain the adjusted first-order and second-order classification codes; finally, the adjusted first-order and second-order classification codes are adaptively weighted to output the pain index of the patient.

[0027] Preferably, the pain index P index is expressed as:

[0028]

[0029] where n represents the preset pain level; N represents the total number of pain levels; P n represents the probability corresponding to each pain level;

[0030] The pain index P index satisfies P index ∈[0,10]. The larger the value of the pain index P index is, the higher the intensity of pain.

[0031] Preferably, the pre-trained multi-level interactive attention network is deployed based on edge computing to realize real-time evaluation of the patient's pain intensity.

[0032] Preferably, the system further includes:

[0033] A real-time display module for updating and displaying the pain index of the patient in real time, as well as the change trend of the pain index.

[0034] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0035] The present invention provides an intelligent evaluation system for chronic pain intensity based on heart rate variability analysis, including: a data acquisition module for real-time collecting electrocardiogram data of patients and real-time extracting a set of HRV features; a dimensionality reduction module for respectively performing manual screening and PCA dimensionality reduction on the extracted set of HRV features to obtain a first dimensionality-reduced feature set and a second dimensionality-reduced feature set; a feature splicing module for splicing and normalizing the HRV features in the first dimensionality-reduced feature set and the second dimensionality-reduced feature set to obtain spliced features; and a pain evaluation module for inputting the spliced features into a pre-trained multi-level interactive attention network to evaluate the pain intensity level.

[0036] The present invention utilizes the electrocardiogram data of patients and, through statistical analysis methods, PCA dimensionality reduction technology, and multi-level interactive attention network technology, can calculate in real time the pain index represented by the ECG data and display a line graph representing the pain change trend on the interface, realizing a quantitative evaluation of the pain degree of patients; the present invention can provide more efficient, personalized, and dynamic pain supervision services for patients and medical staff in real time. Compared with traditional methods, the present invention not only improves the effect of pain management but also improves the overall medical experience and provides a basis for future medical research and technological development. Brief Description of the Drawings

[0037] Figure 1 It is a structural diagram of an intelligent evaluation system for chronic pain intensity based on heart rate variability analysis provided in Embodiment 1.

[0038] Figure 2 It is a structural diagram of an intelligent evaluation system for chronic pain intensity based on heart rate variability analysis provided in Embodiment 2.

[0039] Figure 3 It is a schematic diagram of electrocardiogram acquisition sites provided in Embodiment 2.

[0040] Figure 4 It is a flowchart of electrocardiogram data acquisition of subjects provided in Embodiment 2.

[0041] Figure 5 It is a structural diagram of a multi-level interactive attention network provided in Embodiment 2.

[0042] Figure 6 It is a structural diagram of an interactive attention fusion unit provided in Embodiment 2.

[0043] Figure 7 It is a comparison chart of experimental results with different PCA dimensionality reduction quantities provided in Embodiment 2. Detailed Embodiments

[0044] The drawings are only for illustrative purposes and should not be construed as a limitation of this patent;

[0045] To better illustrate this embodiment, some components in the drawings are omitted, enlarged or reduced, which do not represent the dimensions of the actual product;

[0046] For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0047] The technical solution of the present invention will be further described below in conjunction with the drawings and embodiments.

[0048] Embodiment 1

[0049] As Figure 1 shown, this embodiment provides an intelligent chronic pain intensity assessment system based on heart rate variability analysis, including:

[0050] Data acquisition module 101: used to collect the electrocardiogram data of patients in real time and extract the HRV feature set in real time, and the HRV feature set includes several HRV features;

[0051] Dimensionality reduction module 102: used to perform manual screening and PCA dimensionality reduction on the extracted HRV feature set respectively to obtain the first dimensionality reduction feature set and the second dimensionality reduction feature set;

[0052] Feature splicing module 103: used to splice and normalize the HRV features in the first dimensionality reduction feature set and the second dimensionality reduction feature set to obtain the spliced features;

[0053] Pain assessment module 104: used to input the spliced features into a pre-trained multi-level interactive attention network for pain intensity level assessment, and the multi-level interactive attention network outputs the pain index of the patient in real time to realize real-time intelligent assessment of chronic pain intensity.

[0054] In the specific implementation process, first, the data acquisition module 101 collects the electrocardiogram data of patients in real time and extracts the HRV feature set in real time, and the HRV feature set includes several HRV features;

[0055] Then, the dimensionality reduction module 102 performs manual screening and PCA dimensionality reduction on the extracted HRV feature set respectively to obtain the first dimensionality reduction feature set and the second dimensionality reduction feature set;

[0056] In this embodiment, manual screening can select the feature set most relevant to pain according to expert experience and knowledge, while PCA dimensionality reduction combines the information extraction of the full feature set and extracts the main feature indicators related to pain;

[0057] Principal component analysis is an unsupervised machine learning algorithm and a technique for exploring the structure of high-dimensional data. It is mainly used for dimensionality reduction of data. Through dimensionality reduction, features that are easier for humans to understand can be discovered, the processing speed of valuable information in samples can be accelerated. In addition, it can also be applied to visualization (reducing to two dimensions) and denoising. The basic principle is to project the original data onto a new coordinate system through a linear transformation so that the variance of the projected data is maximized. The goal of the PCA algorithm is that the amount of information lost in the dimensionality-reduced data should be as small as possible. Through PCA technology, multi-dimensional HRV features are mapped to a low-dimensional space. This low-dimensional space consists of brand-new orthogonal features, which are low-dimensional features reconstructed based on the original multi-dimensional features.

[0058] After that, the feature splicing module 103 splices and normalizes the HRV features in the first dimensionality-reduced feature set and the second dimensionality-reduced feature set to obtain spliced features.

[0059] Finally, the pain assessment module 104 inputs the spliced features into a pre-trained multi-level interactive attention network to evaluate the pain intensity level. The multi-level interactive attention network outputs the patient's pain index in real time, realizing the real-time intelligent assessment of chronic pain intensity.

[0060] In this embodiment, the multi-level interactive attention network is designed based on cross-attention and is used to calculate the pain index from HRV features at different first-order and second-order levels. The network simultaneously inputs the spliced features obtained by statistical analysis and PCA technology, and the second-order features obtained by calculating the outer product of the spliced features. The first-order features can be used to capture the importance of individual features, while the second-order features are used to focus on the relationships between features. In the intermediate stage, the model will perform classification coding fusion on the first-order and second-order features, using information at different levels to assist the learning of hierarchical classification coding, and perform adaptive weighting on the multi-level features in the final stage, so as to obtain a pain index that takes into account both the importance of each HRV feature and the correlation between HRV features.

[0061] By studying the relationship between pain and electrocardiogram, this system can help patients and medical staff better understand the pain mechanism and improve pain treatment methods. This system is based on principal component analysis (PCA) and a multi-level interactive attention network, and calculates the pain score representing the pain degree in real time through HRV indicators, thereby assisting doctors in intuitively quantifying the pain degree of patients.

[0062] Embodiment 2

[0063] As Figure 2 shown, this embodiment provides an intelligent chronic pain intensity assessment system based on heart rate variability analysis, including:

[0064] Data acquisition module 201: It is used to collect the electrocardiogram data of patients in real time and extract the HRV feature set in real time. The HRV feature set includes several HRV features;

[0065] Dimensionality reduction module 202: It is used to perform manual screening and PCA dimensionality reduction on the extracted HRV feature set respectively to obtain the first dimensionality reduction feature set and the second dimensionality reduction feature set;

[0066] Feature splicing module 203: It is used to splice and normalize the HRV features in the first dimensionality reduction feature set and the second dimensionality reduction feature set to obtain the spliced features;

[0067] Pain assessment module 204: It is used to input the spliced features into the pre-trained multi-level interactive attention network for pain intensity level assessment. The multi-level interactive attention network outputs the pain index of the patient in real time to realize the real-time intelligent assessment of chronic pain intensity;

[0068] Real-time display module 205, which is used to update and display the pain index of the patient in real time, as well as the change trend of the pain index;

[0069] In the data acquisition module 201, the electrocardiogram data of patients is collected in real time by the bipolar lead method of RA and LL positions;

[0070] The collected electrocardiogram data is denoised and R peak located to obtain the RR sequence data of the electrocardiogram data, and further calculate several features of HRV in the time domain, frequency domain and non-linear domain, and extract the HRV feature set;

[0071] The HRV features at least include: mean heart period, standard deviation of heart period, standard deviation of difference between consecutive heart periods, root mean square of difference between consecutive heart periods, median of heart period, logarithm of difference between consecutive heart periods greater than 50ms, percentage of NN50 in total heart period logarithm, logarithm of difference between consecutive heart periods greater than 20ms, percentage of NN20 in total heart period logarithm, difference between maximum and minimum of heart period, coefficient of variation of difference between consecutive heart periods, coefficient of variation of heart period, average heart rate, maximum heart rate, minimum heart rate, standard deviation of heart rate, total power, very low frequency power with frequency range of 0.0033 - 0.04Hz, low frequency power with frequency range of 0.04 - 0.15Hz, high frequency power with frequency range of 0.15 - 0.4Hz, ratio of low frequency to high frequency power, relative proportion of low frequency power in total power, relative proportion of high frequency power in total power, and regression slope of RR sequence;

[0072] In the dimensionality reduction module 202, several significant HRV features are manually selected from the HRV feature set to obtain the first dimensionality reduction feature set;

[0073] Using the PCA dimensionality reduction algorithm, all HRV features are projected onto the orthogonal coordinate system with the largest variance through linear transformation, several principal component features are retained, and a second dimensionality reduction feature set is obtained;

[0074] The first dimensionality reduction feature set includes the following 8 significant HRV features:

[0075] Standard deviation of the heart rate intervals, root mean square of the differences between consecutive heart rate intervals, coefficient of variation of the heart rate intervals, average heart rate, high-frequency power in the frequency range of 0.15 - 0.4 Hz, low-frequency to high-frequency power ratio, relative proportion of high-frequency power in the total power, and regression slope of the RR sequence;

[0076] The PCA dimensionality reduction algorithm includes:

[0077] Perform standardization processing on all HRV features, calculate the covariance matrix of all standardized HRV features, perform eigenvalue decomposition on the covariance matrix to obtain several eigenvalues and their corresponding eigenvectors, where the eigenvalue represents the variance of each principal component, and the eigenvector represents the direction of the principal component;

[0078] Arrange the eigenvalues in descending order, and the corresponding eigenvectors are also arranged in the same order; the eigenvector with a larger eigenvalue represents a more important principal component;

[0079] Select the first 8 eigenvectors according to the size of the eigenvalues and construct a projection matrix;

[0080] Use the projection matrix to project all HRV features into the principal component space, obtain the dimensionality-reduced HRV features, and jointly save the dimensionality-reduced HRV features as the second dimensionality reduction feature set;

[0081] In the pain assessment module 204, in the pain assessment module, the spliced features are used as first-order features, the first-order features are subjected to outer product calculation to obtain second-order features, and each element in the second-order features represents the relationship between sample points;

[0082] Input the first-order features and the second-order features into a pre-trained multi-level interactive attention network, and a Transformer encoder and an interactive attention fusion unit are set in the multi-level interactive attention network;

[0083] The first-order features and second-order features are first calculated by two groups (1 first-order feature and 4 second-order features) of independent Transformer encoders to obtain the feature encoding and classification encoding of each order. Subsequently, the obtained feature encoding and classification encoding of each order are jointly input into an Interactive Attention fusion unit for multi-level interactive fusion. In the Interactive Attention fusion unit, the classification encoding of each order is respectively subjected to mutual attention calculation with the feature encoding of different orders to obtain the adjusted first-order and second-order classification encodings. Taking the first order as an example, the classification encoding of the first order will perform mutual attention calculation in the Interactive Attention fusion module in combination with the feature encoding of the second order to obtain the adjusted first-order classification encoding through learning. Finally, the adjusted first-order and second-order classification encodings are adaptively weighted to obtain the pain index of the patient.

[0084] The pain index P index is expressed as:

[0085]

[0086] where n represents the preset pain level; N represents the total number of pain levels; P n represents the probability corresponding to each pain level;

[0087] The pain index P index satisfies P index ∈[0, 10], and the larger the value of the pain index P index is, the higher the pain intensity represents;

[0088] The pre-trained multi-level interactive attention network is deployed based on edge computing to realize real-time evaluation of the patient's pain intensity.

[0089] In the specific implementation process, starting from the changes in pain and HRV index features, this system collects the electrocardiogram data of the subjects before and after pain induction and before and after pain treatment, as well as the electrocardiogram data of long-term chronic pain patients such as low back pain, shoulder pain, and limb muscle pain before and after treatment, extracts and analyzes the HRV feature indicators, and finds the evaluation of the pain index, an indicator related to pain changes.

[0090] First, the data acquisition module 201 collects the electrocardiogram data of the patient in real time and extracts the HRV feature set in real time. The HRV feature set includes several HRV features.

[0091] Specifically, with the consent of the patient, the doctor selects eligible long-term chronic pain patients such as low back pain, shoulder pain, and limb muscle pain for the experiment; under the guidance of the doctor, uses a transcutaneous electrical stimulation therapeutic apparatus to treat the patient, and collects the electrocardiogram data of the patient before and after treatment. The collection sites are asFigure 3 As shown, the RA and LL bipolar leads were connected to obtain the patient's subjective NRS score;

[0092] The collected ECG data are denoised and R peaks are located to obtain the RR sequence data of each ECG data, and 24 HRV indicators in the time domain, frequency domain and nonlinear domain are calculated, namely: time domain (Mean_nni, Sdnn, Sdsd, Rmssd, Median_nni, Nni_50, Pni_50, Nni_20, Pni_20, Range_nni, Cvsd, Cvnni, Mean_hr, Max_hr, Min_hr, Std_hr), frequency domain (Total_power, VLF, LF, HF, lf_hf_ratio, lfnu, hfnu), nonlinear domain (Slope); the specific meaning of each indicator is shown in Table 1:

[0093] Table 1 Meaning of HRV index and corresponding index

[0094]

[0095]

[0096] Paired sample T test was performed on the 24 HRV indices of patients before and after treatment to analyze the changes of the 24 HRV indices before and after treatment;

[0097] In addition, this embodiment also recruited 38 university student volunteers to conduct pain induction treatment experiments to verify the relationship between pain and HRV indicators;

[0098] Volunteers must meet the following requirements:

[0099] (1) Aged between 20 and 28 years old;

[0100] (2) no history of chronic pain;

[0101] (3) no recent use of neurological-related drugs;

[0102] (4) Good health and no cardiovascular disease;

[0103] Finally, 38 subjects were selected, including 20 males and 18 females;

[0104] To ensure that the subjects were not under psychological stress, we communicated clearly with them before the experiment and conducted an allergy test using capsaicin. The patients were clearly informed that they would be stimulated by capsaicin during the subsequent experiment, which reduced the stress response of the subjects.

[0105] like Figure 4For the experimental process shown, electrocardiogram data of the subjects before and after pain induction and before and after pain treatment were collected, and a numerical rating scale NRS form was obtained for each stage; capsaicin was selected as the pain induction method; a total of 400 mg of capsaicin with a diameter of 4 cm was applied to the calf of the left leg of the subject; and a hot air blower was used to perform heat treatment on the capsaicin-stimulated area to accelerate pain induction; after the induction time reached 25 minutes, the subjective NRS pain score of the subject was asked every minute until the subject's feedback score reached 7 points or the time length reached half an hour.

[0106] For the method of treating pain, under the guidance of a doctor, Chengfu acupoint (International label: BL36) and Yinmen acupoint (International label: BL37) were selected, and transcutaneous electrical stimulation therapy instrument was used for electrical stimulation.

[0107] According to the data processing method described above, 24 indexes of HRV in the time domain, frequency domain and non-linear domain were obtained.

[0108] Finally, the HRV features extracted from all patients and subjects were summarized as the HRV feature set.

[0109] Then, the dimensionality reduction module 202 respectively performed manual screening and PCA dimensionality reduction on the extracted HRV feature set to obtain the first dimensionality reduction feature set and the second dimensionality reduction feature set.

[0110] Based on the above 24 indexes of HRV extracted, the significant changes in HRV indexes and NRS score changes of the patients before and after treatment were analyzed; the significant changes in HRV indexes between the resting pain-free state and the capsaicin-induced pain state of the subjects, the significant changes in HRV indexes between the resting pain-free state and after treatment, the NRS score changes between the resting pain-free state and the capsaicin-induced pain state, and the NRS score changes between the resting pain-free state and after treatment were analyzed.

[0111] Combined with the real patient data for analysis, the following 8 HRV indexes of the subjects and patients were selected as the first dimensionality reduction feature set: Sdnn: standard deviation of all RR intervals, Rmssd: root mean square of adjacent RR intervals, Cvnni: ratio of the standard deviation of the differences between adjacent RR intervals to the mean of the RR intervals, Mean_hr: mean heart rate, HF: high-frequency power component in HRV, Lf_hf_ratio: ratio of low-frequency power component to high-frequency power component in HRV, Hfnu: relative proportion of HF component in the total power of HRV, Slope: linear regression slope of RR intervals.

[0112] Based on the above 24 indexes of HRV extracted, principal component analysis (PCA) was used to reduce the data dimension and retain the main information, reducing it to 8 dimensions to obtain the second dimensionality reduction feature set.

[0113] In this embodiment, the principal component analysis method is an unsupervised machine learning algorithm, a technique for exploring the structure of high-dimensional data, mainly used for dimensionality reduction of data. Through dimensionality reduction, features that are easier for humans to understand can be discovered, the processing speed of valuable information in samples can be accelerated. In addition, it can also be applied to visualization (reducing to two dimensions) and denoising. The basic principle is to project the original data onto a new coordinate system through a linear transformation so that the variance of the projected data is maximized. The goal that the PCA algorithm aims to achieve is that the information loss of the data after dimensionality reduction should be as small as possible. Through the PCA technique, multi-dimensional HRV features are mapped onto a low-dimensional space. This low-dimensional space consists of brand-new orthogonal features, which are low-dimensional features reconstructed based on the original multi-dimensional features.

[0114] After that, the feature splicing module 203 splices and normalizes the HRV features in the first dimensionality reduction feature set and the second dimensionality reduction feature set to obtain spliced features.

[0115] Specifically, the 8-dimensional HRV indicators selected manually are spliced with the 8-dimensional HRV indicators projected from 24 HRV indicators synthesized using the PCA technique. In this way, it is possible to focus on the 8 indicators that have been found to have more obvious changes in the significance analysis, and at the same time utilize the 8 indicators obtained through the PCA technique that take into account the overall 24 indicators. Subsequently, mean-variance normalization is used for standardization processing. The formula is as follows:

[0116]

[0117] where x is a single 16-dimensional sample after splicing, μ is the mean of the corresponding sample, S is the variance of the corresponding sample, and X std is the standardized sample. Through standardization, each sample is processed so that the data distribution has a mean equal to 0 and a variance equal to 1, avoiding the influence of large data on the entire sample.

[0118] Finally, the pain assessment module 204 inputs the spliced features into a pre-trained multi-level interactive attention network for pain intensity level assessment. The multi-level interactive attention network outputs the pain index of the patient in real time, realizing real-time intelligent assessment of chronic pain intensity.

[0119] In this embodiment, a multi-level interactive attention network is designed in this module to evaluate pain based on HRV metrics; the multi-level interactive attention network is pre-trained using the data in Part A of the publicly available dataset, the BioVid Heat Pain Database; the ECG data in Part A of this database was measured by bipolar leads at the upper right and lower left corners of the body, including 90 participants in three age groups of 30 subjects each aged 18 - 35, 36 - 50, and 51 - 65, with each age group evenly divided into 15 male and 15 female subjects; due to technical problems during the recording process, the data of three subjects was missing, so a total of 87 participants' data was available; for each subject, there were five different stimulus levels, with 20 samples for each level, each sample containing 5.5 seconds of signal, sampled at a frequency of 512 Hz, for a total of 2816 sample points; the entire dataset totaled 8700 data points, divided into five pain levels, with 1740 data points for each level;

[0120] Using the 8700 data points, the standardized 16-dimensional data for each 5.5-second sample was extracted according to the previous steps, and four different binary classification tasks were performed, namely (P0 VS P1), (P0 VS P2), (P0 VS P3), and (P0 VS P4), which represent the pain threshold assessment task, mild pain assessment task, severe pain assessment task, and pain tolerance assessment task respectively; through these four tasks, four models with the same structure but different parameters can be constructed to analyze the pain situation of the human body; finally, the pre-trained multi-level interactive attention network was obtained, and the results of the pre-trained model are as Figure 5 shown, and its structure consists of a Transformer encoder and an interactive attention fusion unit (the structure is as Figure 6 shown); in this embodiment, when using the multi-level interactive attention network to input 16-dimensional data for classifying four-level pain and no pain, an accuracy of 67.28% can be achieved, and the changes in HRV metrics vary with different pain levels, and the degree of variation is also different, resulting in inconsistent discrimination;

[0121] Secondly, in this embodiment, in order to verify that splicing the 24-dimensional HRV metrics after dimensionality reduction by PCA with the selected 8-dimensional features can improve the accuracy, an additional set of experiments was conducted; the painless data and extreme pain data in the heat pain dataset were used for verification, and experiments were carried out without splicing the features after PCA dimensionality reduction, PCA reduced to 6 dimensions, PCA reduced to 8 dimensions, and PCA reduced to 10 dimensions, and the experimental results are as Figure 7 shown;

[0122] Figure 7The experimental results using different numbers of PCA dimensionality reductions are shown. When the features after PCA dimensionality reduction are not concatenated, the results are the worst; while when PCA is reduced to 8 dimensions, the results are the best. Therefore, when calculating the pain index later, PCA is reduced to 8 dimensions for use;

[0123] To verify the effectiveness of the multi-level interactive attention network, this embodiment also selects different baseline methods for comparison. These methods use the same dataset as the system and are all trained using the leave-one-subject method; The experimental results are shown in Table 2:

[0124]

[0125]

[0126] As can be seen from Table 2, among the existing methods, the best result is the network framework improved based on the transformer used by the multimodal automatic assessment framework, with an accuracy rate reaching 67.04%, and the number of parameters and floating-point operations reaching 6.03M and 1.25G respectively; while in the context of the same dataset and the same training strategy, the accuracy rate of this system reaches 67.28%, exceeding the existing best method;

[0127] In addition, considering the real-time requirement for calculating the pain level and the limited computing power of edge devices, it is necessary to control the size of the model's parameters and floating-point operations; The method of this system only uses 16 features to classify the pain level, greatly reducing the number of model parameters and floating-point operations, which are only 19.52K and 237.18K respectively, reducing the number of model parameters by about 300 times and the number of model floating-point operations by about 5000 times compared to the existing best network; Deploying this method on the edge computing unit can achieve real-time assessment of the patient's pain level;

[0128] Finally, the real-time display module 205 is used to update and display the patient's pain index in real time, as well as the change trend of the pain index;

[0129] Based on the data analysis in the previous steps, a 5.5-second ECG signal can be obtained and the corresponding pain trend can be calculated, and this trend is displayed on the interface in real time; The display logic is as follows: At the beginning of the acquisition, data is input into the constructed network in real time every 5.5 seconds to obtain the probability values of the corresponding four pain categories, and the expected value of the pain level is obtained as the pain index, specifically as the following formula:

[0130]

[0131] Among them, n ∈ [1, 4] represents the four pain levels from the beginning of feeling pain to extreme pain, Pn represents the probability corresponding to each pain level, P index ∈[0, 10] represents the pain expectation value of the current object. The larger the value, the higher the degree of pain;

[0132] After obtaining the corresponding pain expectation value (i.e., pain index) every 5.5 seconds, a point is drawn on the display interface and displayed on the interface in real time, thereby showing the change trend of the pain score caused by pain in real time;

[0133] This system uses the electrocardiogram data of patients, subjects and public data sets, and through statistical analysis methods, PCA technology, multi-level interactive attention network technology, calculates in real time the pain index characterized by the ECG data every 5.5 seconds, and displays a line graph representing the pain change trend on the interface; enabling doctors to quantitatively evaluate the pain level of patients by seeing the pain change trend of patients from the interface in real time.

[0134] The same or similar reference numerals correspond to the same or similar components;

[0135] The terms describing the positional relationship in the drawings are for illustrative purposes only and should not be construed as a limitation of this patent;

[0136] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention and are not intended to limit the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all the implementation manners here. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the claims of the present invention.

Claims

1. An intelligent assessment system for chronic pain intensity based on heart rate variability analysis, characterized in that: include: Data acquisition module: used to collect the patient's electrocardiogram data in real time and extract the HRV feature set in real time, wherein the HRV feature set includes several HRV features; Dimensionality reduction module: used to perform manual screening and PCA dimensionality reduction on the extracted HRV feature set to obtain a first dimensionality reduction feature set and a second dimensionality reduction feature set; A feature splicing module: used for splicing and normalizing the HRV features in the first dimensionality reduction feature set and the second dimensionality reduction feature set to obtain a splicing feature; Pain assessment module: used to input the splicing features into a pre-trained multi-level interactive attention network to assess the pain intensity level. The multi-level interactive attention network outputs the patient's pain index in real time to achieve real-time intelligent assessment of chronic pain intensity.

2. The intelligent chronic pain intensity assessment system based on heart rate variability analysis according to claim 1, characterized in that: In the data acquisition module, the patient's electrocardiogram data is collected in real time by means of bipolar leads at the RA and LL points; The collected electrocardiogram data is denoised and R peaks are located to obtain RR sequence data of the electrocardiogram data, and several features of HRV in the time domain, frequency domain and nonlinear domain are further calculated to extract the HRV feature set.

3. The intelligent chronic pain intensity assessment system based on heart rate variability analysis according to claim 1, characterized in that: The HRV characteristics include at least: average heartbeat interval, standard deviation of heartbeat interval, standard deviation of continuous heartbeat interval difference, root mean square of continuous heartbeat interval difference, median of heartbeat interval, logarithm of continuous heartbeat interval difference greater than 50ms, percentage of NN50 in total heartbeat interval logarithm, logarithm of continuous heartbeat interval difference greater than 20ms, percentage of NN20 in total heartbeat interval logarithm, difference between maximum and minimum heartbeat interval, coefficient of variation of continuous heartbeat interval difference, coefficient of variation of heartbeat interval, average heart rate, maximum heart rate, minimum heart rate, standard deviation of heart rate, total power, extremely low frequency power in the frequency range of 0.0033-0.04Hz, low frequency power in the frequency range of 0.04-0.15Hz, high frequency power in the frequency range of 0.15-0.4Hz, ratio of low frequency to high frequency power, relative proportion of low frequency power in total power, relative proportion of high frequency power in total power and regression slope of RR sequence.

4. The intelligent chronic pain intensity assessment system based on heart rate variability analysis according to claim 3, characterized in that: In the dimension reduction module, a number of significant HRV features are manually screened out from the HRV feature set to obtain a first dimension reduction feature set; Using the PCA dimensionality reduction algorithm, all HRV features are projected to the orthogonal coordinate system with the largest variance through linear transformation, and several principal component features are retained to obtain the second dimensionality reduction feature set.

5. The intelligent chronic pain intensity assessment system based on heart rate variability analysis according to claim 4, characterized in that: The first dimension reduction feature set includes the following 8 significant HRV features: The standard deviation of heart beat intervals, the root mean square of the difference between consecutive heart beat intervals, the coefficient of variation of heart beat intervals, the average heart rate, the high-frequency power in the frequency range of 0.15-0.4Hz, the ratio of low-frequency to high-frequency power, the relative proportion of high-frequency power in the total power and the regression slope of the RR series.

6. The intelligent chronic pain intensity assessment system based on heart rate variability analysis according to claim 4, characterized in that: The PCA dimensionality reduction algorithm includes: Standardize all HRV features, calculate the covariance matrix of all standardized HRV features, perform eigenvalue decomposition on the covariance matrix to obtain a number of eigenvalues ​​and their corresponding eigenvectors, wherein the eigenvalue represents the variance of each principal component and the eigenvector represents the direction of the principal component; Arrange the eigenvalues ​​from large to small, and the corresponding eigenvectors in the same order; the eigenvector with a larger eigenvalue represents a more important principal component; Select the first 8 eigenvectors according to the size of the eigenvalue and construct the projection matrix; All HRV features are projected into the principal component space using the projection matrix to obtain HRV features after dimensionality reduction, and the HRV features after dimensionality reduction are collectively saved as a second dimensionality reduction feature set.

7. The intelligent chronic pain intensity assessment system based on heart rate variability analysis according to claim 1, characterized in that: In the pain assessment module, the concatenated features are used as first-order features, and the first-order features are calculated by outer product to obtain second-order features, and each element in the second-order features represents the relationship between sample points; Inputting the first-order features and the second-order features into a pre-trained multi-level interactive attention network, wherein the multi-level interactive attention network is provided with a Transformer encoder and an interactive attention fusion unit; The first-order features and second-order features are first calculated by two groups of independent Transformer encoders to obtain feature codes and classification codes of each order; then the obtained feature codes and classification codes of each order are input into the interactive attention fusion unit for multi-level interactive fusion; in the interactive attention fusion unit, the classification codes of each order are respectively subjected to mutual attention calculation with feature codes of different orders to obtain adjusted first-order and second-order classification codes; finally, the adjusted first-order and second-order classification codes are adaptively weighted to output the patient's pain index.

8. The intelligent chronic pain intensity assessment system based on heart rate variability analysis according to claim 7, characterized in that: The pain index P index It is expressed as: Where n represents the preset pain level; N represents the total number of pain levels; P n Indicates the probability corresponding to each pain level; The pain index P index Satisfy P index ∈[0,10], pain index P index The larger the value, the higher the intensity of the pain.

9. The intelligent chronic pain intensity assessment system based on heart rate variability analysis according to claim 1, characterized in that: The pre-trained multi-level interactive attention network is deployed based on edge computing to achieve real-time assessment of the patient's pain intensity.

10. A chronic pain intensity intelligent assessment system based on heart rate variability analysis according to any one of claims 1 to 9, characterized in that: The system further comprises: The real-time display module is used to update and display the patient's pain index and the changing trend of the pain index in real time.

Images