Search

CN-122000064-A - Pain discrimination method, system, equipment and storage medium based on time sequence and multiple modes

CN122000064ACN 122000064 ACN122000064 ACN 122000064ACN-122000064-A

Abstract

The application provides a pain judging method based on time sequence and multiple modes, which comprises the steps of collecting conventional physiological indexes, behavior characteristic indexes and potential auxiliary indexes, preprocessing the multiple mode data, including abnormal value processing, individual normalization and time alignment, constructing a time sequence data block, establishing an individual base line, dynamically calibrating a base line threshold every 7 days, calibrating a blood sugar base line every 3 days for diabetics, extracting the physiological characteristics, the behavior characteristic and the time sequence characteristics of the time sequence data block through an AI model, fusing the multiple mode characteristics through an attention mechanism, synchronously verifying the multiple mode indexes, calculating multiple classification probability and judging the classification threshold, identifying full-level pain, outputting a clinical intervention strategy corresponding to pain level, dynamically updating model parameters and the base line threshold based on clinical intervention feedback and newly added crowd data.

Inventors

  • Xie Dongjing
  • Kong Daying
  • DU YIHUA

Assignees

  • 无锡市人民医院

Dates

Publication Date
20260508
Application Date
20260129

Claims (10)

  1. 1. A pain discrimination method based on time sequence and multiple modes, which is characterized by comprising the following steps: s1, acquiring multi-modal data, wherein the multi-modal data comprises conventional physiological indexes, behavioral characteristic indexes and potential auxiliary indexes; s2, preprocessing the multi-mode data, including outlier processing, individual normalization and time alignment, and constructing a time sequence data block; s3, establishing an individual baseline based on the rest state data of 3 days before the patient, dynamically calibrating a baseline threshold every 7 days, and calibrating a blood sugar baseline of 1 time every 3 days for the diabetic patient; S4, extracting physiological characteristics, behavioral characteristics and time sequence characteristics of the time sequence data block through an AI model, and fusing multi-mode characteristics by adopting a transducer attention mechanism, wherein a fusion formula is as follows: +β +γ +δ ; Wherein the method comprises the steps of In order to fuse the features of the features, As a result of the conventional physiological characteristics, As a feature of the behavior of the vehicle, As a potential assist feature, For the time sequence feature, α is the weight of the normal physiological feature, β is the weight of the behavioral feature, γ is the weight of the potential auxiliary feature, δ is the weight of the time sequence feature, α+β+γ+δ=1; s5, identifying NRS 0-10 score full-level pain through multi-mode index synchronous verification, multi-classification probability calculation and classification threshold judgment; and S6, outputting a clinical intervention strategy corresponding to the pain level, and dynamically updating model parameters and a baseline threshold value based on clinical intervention feedback and newly-added crowd data.
  2. 2. The method of claim 1, wherein the normal physiological index comprises cardiovascular system index, neuromuscular system index, nervous system index, respiratory system index and endocrine system index; the cardiovascular system indicators include heart rate, systolic pressure, and heart rate variability; the neuromuscular system index includes a galvanic skin response and myoelectricity; The nervous system index comprises brain electricity and pupil diameter; The respiratory index comprises respiratory frequency and the endocrine index comprises cortisol.
  3. 3. The method of claim 1, wherein the behavioral indicators include facial expressions, limb movements, and sound characteristics; The facial expression is collected through a camera and identified by adopting a CNN convolutional neural network; The limb actions are collected through the combination of the gesture sensor and the camera, and are identified by adopting a skeleton key point detection algorithm; the sound features are collected through a microphone, and the fundamental frequency F0 and the frequency spectrum entropy value features are extracted.
  4. 4. The method of claim 1, wherein the potential auxiliary indicators include blood glucose, skin microcirculation blood flow velocity and alpha-amylase activity; The blood sugar is collected by a noninvasive near infrared spectrum sensor; the skin microcirculation blood flow velocity is acquired by a wrist type laser Doppler sensor; The alpha-amylase activity was collected by saliva test paper.
  5. 5. The method for discriminating pain based on time sequence and multiple modes according to claim 1 wherein in said step S2, physiological index extreme values are removed by adopting a3σ principle, and behavioral characteristic abnormal value processing is realized by filtering environmental interference by an interframe difference method; the individual normalization is through the formula Calculating the change rate of each index, wherein R is the change rate, The values are collected in real time for the index, Individual baseline values are indicators; the time alignment binds the multi-mode data through the time stamp, and a time sequence data block is constructed according to the unit of 5 minutes.
  6. 6. The method for determining pain based on time sequence and multiple modes according to claim 1, wherein step S4 extracts physiological features, behavioral features and time sequence features of the time sequence data block through an AI model; The adoption of a transducer attention mechanism to fuse the multi-mode features comprises physiological feature extraction, behavior feature extraction, time sequence feature extraction and attention fusion; The physiological characteristic extraction is used for extracting time domain characteristics and frequency domain characteristics based on a CNN convolutional neural network; The behavior feature extraction is performed, facial expression classification features are extracted through a CNN convolutional neural network, limb action amplitude classification features are extracted through a gesture estimation algorithm, and classification differences of fundamental frequency changes are extracted through a voice algorithm; the time sequence feature extraction is realized by capturing the index change trend and peak synchronism of 25 minutes continuously by adopting an LSTM long-term memory network; The attention fusion adopts a transducer attention mechanism to distribute modal weights, pain levels are divided into mild pain, moderate pain and severe pain, the multi-modal weights are dynamically adjusted according to different pain levels, wherein the weight distribution of different pain levels meets the condition that the mild pain behavior characteristic weight is more than or equal to the conventional physiological characteristic weight, the severe pain conventional physiological characteristic weight is more than or equal to 0.5, and the special crowd behavior characteristic weight is 0.
  7. 7. The method of claim 1, wherein the full-level pain identification logic of step S5 comprises: Synchronously verifying the multi-mode indexes, and judging whether the change of the core indexes accords with the unidirectional trend and the amplitude interval of the pain of the corresponding level; Calculating multi-classification probability, and outputting probability distribution of each grade of NRS 0-10 through LSTM, transformer and Softmax combined model; Judging a grading threshold, presetting a grading probability threshold, and triggering secondary verification when the grading probability threshold is lower than the grading probability threshold; and an anti-interference mechanism for automatically reducing the modal weight when the single modal index is abnormal.
  8. 8. A time series and multi-modal based pain discrimination system, comprising: The multi-mode data acquisition module is integrated with conventional physiological index acquisition equipment, behavior characteristic acquisition equipment and potential auxiliary index acquisition equipment and is used for acquiring multi-mode data; the data preprocessing module is used for executing abnormal value rejection, individual normalization and time alignment operations; the baseline calibration module is used for establishing and dynamically calibrating individual baseline thresholds; the AI feature fusion module is used for extracting physiological features, behavior features and time sequence features of the time sequence data block and fusing multi-mode features through an attention mechanism; the full-level pain identification module is used for executing NRS 0-10 score full-level pain identification through multi-mode index synchronous verification, multi-classification probability calculation and classification threshold judgment; The hierarchical intervention output module is used for outputting a clinical intervention strategy; and the model updating module is used for dynamically updating model parameters based on clinical intervention feedback and newly-added crowd data.
  9. 9. A computer device comprising a processor, a memory and a program or instruction stored on the memory and executable on the processor, the program or instruction when executed by the processor implementing the steps of the method according to any one of claims 1 to 7.
  10. 10. A computer readable storage medium, characterized in that the storage medium has stored thereon a program or instructions which, when executed by a processor, implement the steps of the method according to any of claims 1 to 7.

Description

Pain discrimination method, system, equipment and storage medium based on time sequence and multiple modes Technical Field The application relates to the technical field of medical monitoring, in particular to a pain judging method, a pain judging system, pain judging equipment and a pain judging storage medium based on time sequence and multiple modes. Background The digital grading scale (NRS) is a quantitative method for comprehensively evaluating the severity of nursing, can effectively help medical institutions evaluate the severity of nursing patients, provides valuable decision support for nursing teams, and is one of the most commonly used evaluation tools in the field of nursing. Pain is an important stress reaction of a human body, particularly severe pain of NRS 8-10 minutes, and complications such as sudden rise of blood pressure, nausea and vomiting, exacerbation of tissue injury, anxiety and depression and the like are extremely easy to cause if the pain is not recognized and intervened in time, so that prognosis and quality of life of a patient are seriously influenced. The traditional monitoring mode takes active description of patients or visual evaluation of medical staff as a core, special crowds such as coma, infants, cognitive impairment and the like which cannot express pain cannot be covered, the hidden pain recognition rate of the crowds is low, the illness state is easy to deteriorate because the pain is not perceived, the traditional mode excessively depends on subjective reports, and the coverage range is limited. In addition, most of the prior art focuses on modeling of single physiological indexes (such as heart rate and skin electricity), is easily interfered by irrelevant factors such as exercise, diet, environmental temperature and the like, has the recognition accuracy of severe pain of only 62% -75%, is difficult to distinguish pain stress and other physiological stress states, has weak anti-interference capability of single index monitoring, and has low recognition accuracy. Meanwhile, physiological differences exist in the monitored objects, the physiological baseline differences (such as resting heart rate and basic blood sugar level) of the individuals are not considered in the traditional mode, the judgment is carried out by adopting a fixed threshold value, misjudgment is easy to be caused, and the method is particularly not suitable for special crowds with fluctuation of physiological indexes such as diabetics and the elderly. Therefore, there is a need to develop a method, system, apparatus and storage medium for determining pain in time series and multiple modes to solve the above-mentioned problems. It should be noted that the foregoing is only used to assist in understanding the technical solution of the present application, and does not represent an admission that the foregoing is prior art. Disclosure of Invention The method, the system, the equipment and the storage medium for judging the pain based on time sequence and multiple modes can cover various crowds, output a hierarchical intervention strategy, shorten intervention delay and improve clinical practicability and accuracy of pain management. In a first aspect, the present application relates to a method for pain discrimination based on time sequence and multiple modes, comprising the steps of: s1, acquiring multi-modal data, wherein the multi-modal data comprises conventional physiological indexes, behavioral characteristic indexes and potential auxiliary indexes; s2, preprocessing the multi-mode data, including outlier processing, individual normalization and time alignment, and constructing a time sequence data block; s3, establishing an individual baseline based on the rest state data of 3 days before the patient, dynamically calibrating a baseline threshold every 7 days, and calibrating a blood sugar baseline of 1 time every 3 days for the diabetic patient; s4, extracting physiological characteristics, behavioral characteristics and time sequence characteristics of the time sequence data block through an AI model, and fusing multi-mode characteristics by adopting an attention mechanism, wherein a fusion formula is as follows: +β+γ+δ; Wherein the method comprises the steps of In order to fuse the features of the features,As a result of the conventional physiological characteristics,As a feature of the behavior of the vehicle,As a potential assist feature,For the time sequence feature, α is the weight of the normal physiological feature, β is the weight of the behavioral feature, γ is the weight of the potential auxiliary feature, δ is the weight of the time sequence feature, α+β+γ+δ=1; s5, identifying NRS 0-10 score full-level pain through multi-mode index synchronous verification, multi-classification probability calculation and classification threshold judgment; and S6, outputting a clinical intervention strategy corresponding to the pain level, and dynamically updating model parameters and a basel