阅读说明：本技术 一种基于置信区间融合阈值判据的疼痛识别系统 (Pain recognition system based on confidence interval fusion threshold criterion ) 是由 张瑜 张露 曹超宇 龚卫娟 朱林 刘林 邹焱 毕雅昕 于 2021-07-12 设计创作，主要内容包括：本发明公开了一种基于置信区间融合阈值判据的疼痛识别系统,包括监测数据采集模块、阈值处理模块、疼痛识别模块、输出模块和报警模块；监测数据采集模块用于采集多维度的监测数据；阈值处理模块用于融合阈值判据对疼痛监测数据进行预处理；疼痛识别模块利用构建的疼痛监测数据分级模型对所述监测数据进行函数训练,确定疼痛检测数据置信区间,输出模块用于对各个实时监测数据训练所对应的疼痛级别范围进行显示,并传送至医护人员电脑；报警模块用于判断监测数据是否在有效范围内,若监测数据无效则发出报警信息提示医护人员进行处理。本发明能对患者疼痛程度进行实时和精确化的监测识别,从而实现患者的有效疼痛评估和处理。(The invention discloses a pain recognition system based on confidence interval fusion threshold criterion, which comprises a monitoring data acquisition module, a threshold processing module, a pain recognition module, an output module and an alarm module, wherein the monitoring data acquisition module is used for acquiring a threshold value; the monitoring data acquisition module is used for acquiring multi-dimensional monitoring data; the threshold processing module is used for fusing threshold criteria to preprocess the pain monitoring data; the pain recognition module performs function training on the monitoring data by using the constructed pain monitoring data grading model to determine a confidence interval of pain detection data, and the output module is used for displaying a pain grade range corresponding to each real-time monitoring data training and transmitting the pain grade range to a medical staff computer; the alarm module is used for judging whether the monitoring data is in an effective range or not, and if the monitoring data is invalid, the alarm module sends alarm information to prompt medical staff to process the monitoring data. The invention can carry out real-time and accurate monitoring and identification on the pain degree of the patient, thereby realizing effective pain evaluation and treatment of the patient.)

1. A pain recognition system based on confidence interval fusion threshold criterion is characterized by comprising a monitoring data acquisition module, a threshold processing module, a pain recognition module, an output module and an alarm module;

the monitoring data acquisition module is used for acquiring multidimensional monitoring data, acquiring vital sign data of a patient through the vital sign detector, acquiring sleep condition data of the patient through the sleep induction device, and inputting pain grading monitoring data of the patient through the medical care information system;

the threshold processing module is used for fusing threshold criteria to preprocess pain monitoring data and carrying out two-step threshold preprocessing on the collected vital sign data, sleep condition data and pain grading monitoring data;

the pain recognition module performs function training on the monitoring data by using the constructed pain monitoring data grading model to obtain a corresponding pain level, determines a confidence interval of pain detection data and obtains a pain monitoring value range of the maximum probability corresponding to a certain pain level;

the output module is used for displaying the pain level range corresponding to each real-time monitoring data training and transmitting the pain level range to the medical staff computer;

the alarm module is used for judging whether the monitoring data are in an effective range or not, and if the monitoring data are invalid, alarm information is sent out to prompt medical staff to process the monitoring data.

2. A pain recognition system based on confidence interval fusion threshold criteria as claimed in claim 1, wherein the vital sign data includes heart rate value, blood pressure value, respiration value, blood oxygen saturation and body temperature of the patient; the sleep condition data includes total sleep time, sleep break time, and deep sleep ratio of the patient.

3. The pain recognition system according to claim 1, wherein the two-step threshold preprocessing specifically comprises:

the method comprises the following steps of firstly, preprocessing, namely comparing all collected monitoring data with a preset threshold value, wherein the range of the preset threshold value is an area between a maximum monitoring threshold value and a minimum monitoring threshold value, and if the monitoring data are within the range of the preset threshold value, namely the numerical value of the monitoring data are within an effective range, further inputting the monitoring data into a constructed pain monitoring data grading model; if the monitoring data at the current moment is not in the range of the normal rest state or the set pain threshold value of the patient, the alarm module sends alarm information to remind medical staff of carrying out corresponding treatment;

and secondly, preprocessing, namely processing the collected vital sign data and the sleep condition data, calculating the variation of the vital sign data and the sleep condition data at a certain moment, comparing the variation with the maximum variation of the monitoring data in the rest state, and if the variation of the monitoring data at a certain moment is larger than the maximum variation in the rest state, indicating that the fluctuation of the vital sign and the sleep condition of the patient is large, further inputting the monitoring data at the moment into the constructed pain monitoring data grading model.

4. A pain recognition system based on confidence interval fusion threshold criteria as claimed in claim 1, wherein the constructed pain monitoring data classification model specifically includes a data input layer, I hidden layers and an output processing layer group.

5. The pain recognition system according to claim 4, wherein the data input layer is configured to input real-time data corresponding to the pain monitoring indicator, including real-time values of the vital sign data, the sleep status data and the pain score, and real-time variation of the vital sign data and real-time variation of the sleep status data.

6. A pain recognition system based on confidence interval fusion threshold criteria as claimed in claim 4, wherein the hidden layer group is set as I hidden layers, and any ith hidden layer group comprises two convolutional layers in turnA pooling layer and a regularization layer, the pain level of layer i being denoted as H_iPain was monitored in the MATLAB program for values 0 to X_tPain level H corresponding to the range_iFitting the parameter function to obtain pain level H under 5 th order limitation_i(X_t) The corresponding best restriction function:

wherein H_i(X_t) Defining an optimal defining function for the pain level at 5 th order; x is the number of_tReal-time specific values for a certain pain monitoring index; alpha is alpha₁，α₂，α₃，α₄，α₅，α₆Best fit coefficients defining a function for order 5; setting the initial dimension of the hidden layer as 100, and then setting the dimension of the hidden layer according to the obtained H_iAnd true experimental value H₀The dimension of the next hidden layer is calculated according to the difference between the first hidden layer and the second hidden layer, and the dimension of the ith hidden layer is expressed as:

7. a pain recognition system based on confidence interval fusion threshold criteria as claimed in claim 4, wherein the set of output processing layers includes a flattening layer, a first fully-connected layer, an I +1 th regularization layer and a second fully-connected layer; the second full-connection layer connects the input feature vectors of the I +1 th regularization layer to the M pain levels of the second full-connection layer, processes the input feature vectors, obtains the feature vectors with one dimension being M, and inputs the feature vectors into the softmax regression classifier, so that the probability of pain monitoring data corresponding to the M pain levels is output, and training is completed.

8. The pain recognition system of claim 4, wherein the confidence interval for determining the confidence interval of the pain detection data is based on a confidence interval fusion threshold criterionThe body includes input vital sign value X_tAnd corresponding pain level H_iEquation f (X) introduced into the MATLAB program under the conditions of the Weibull equation_t) Automatic fitting is carried out, and the Weibull equation mode is obtained as follows:

in the formula, f (X)_t) Monitoring the value X for pain_tExpressed as the number of occurrences or probability of each pain monitoring value in a certain pain level; b₁Is the shape parameter of Weibull equation; b₂Dimension parameters of Weibull equation; b₃Position parameters of Weibull equation; e is the number of fingers; t is numerical numbering of pain monitoring data, f (X) is taken by drawing Weibull probability distribution map_t) The 95% confidence interval is the pain monitoring range of values that represents the maximum probability for a certain pain level.

Technical Field

The invention relates to the technical field of medical treatment, in particular to a pain recognition system based on confidence interval fusion threshold criterion.

Background

Pain is a characteristic of many diseases and can cause pain to patients. When the patient has changed the disease condition, the pain is more intense; when the condition is improved, the pain is weakened; when healing is complete, the pain disappears. Currently, the results of pain assessment are clinically useful as important indicators for monitoring disease and measuring therapeutic efficacy. However, pain is a subjective sensation that is easily disturbed by many factors, such as state of consciousness, mood, sensation, and psychological effects, and, due to its subjectivity, it is an adverse factor for objective assessment of pain sensation and subsequent treatment by the medical staff. The current pain assessment mainly comprises two methods, namely self assessment and observer assessment. The self-assessment method of the patient is convenient and subjective, has good reference significance, is the most widely applied assessment method at present, but the self-assessment can not ensure that each assessment is accurate and credible, and some special groups (such as unconscious patients, aphasia patients, patients in intensive care, patients with dementia, neonates, patients with mental impairment and the like) can not accurately express the pain part and the pain degree of the patient. For such patients, pain is currently assessed and treated mainly by their agents, such as professional medical personnel, parents of infants, etc. However, the method relies on continuous observation and identification of the agent, has low efficiency, is greatly influenced by the subjectivity of the observer, and brings inconvenience to the treatment of medical staff. In addition, improper pain assessment can further lead to improper clinical treatment, and inadequate or excessive use of analgesic drugs can cause physiological and psychological damage to the patient. Therefore, a more objective and accurate method for evaluating the pain level of a patient is needed.

In recent years, along with the development of machine learning and computer technology, automatic pain recognition technology based on emotion recognition and deep learning is more and more emphasized by scholars at home and abroad. At present, the technologies are basically based on the traditional machine learning algorithm, physiological indexes such as vital signs of a patient, even brain waves, electromyographic signals and the like are often collected, wherein the indexes collected by monitoring the vital signs are single, data are not processed, the training method is simple, and errors and false positive conditions often exist in the training process; the acquisition modes of brain waves, electromyograms and the like are relatively complex, and higher equipment and technical requirements are required in actual clinical practice; the pain recognition method based on facial expressions is also more and more concerned by scholars at home and abroad, but due to the complexity and controllability of the facial expressions of adults, the feature points selected on the face are few, the facial conditions and the expressions are more generally divided, and the pain degree subjective felt by a patient is difficult to accurately detect.

However, many studies at home and abroad indicate that vital signs such as Blood Pressure (BP), Respiratory Rate (RR) and Heart Rate (HR) are the most common physiological indicators of pain. Pain produces a physiological stress response, including an increase in heart rate and blood pressure, which provides oxygen and other nutrients to vital organs. These prompt the patient to respond physiologically to pain, such as tachycardia, tachypnea or hypertension, so vital signs can be used as indicators for pain assessment. In addition, pain can affect sleep quality, leading to sleep disturbance; patients with persistent insomnia often experience chronic pain, and the quality of sleep also reflects the pain status of the patient over time. Therefore, collecting multidimensional and versatile monitoring data may more fully reflect the pain level of the patient. In addition, at present, the research on pain identification through data such as vital signs is limited to the pain level corresponding to the numerical value of the monitored data, but the individual difference of the monitored data such as the vital signs is large, and the variation of the data needs to be monitored, so that the pain identification is more individualized and accurate.

Therefore, how to realize accurate, efficient and automatic pain recognition by using technologies such as deep learning and the like so as to evaluate pain patients in real time is an urgent problem to be solved.

Disclosure of Invention

The invention aims to overcome the defects of the prior art and provides a pain recognition system based on confidence interval fusion threshold criterion, which can monitor and recognize the pain degree of a patient in real time and accurately, thereby realizing effective pain evaluation and treatment of the patient.

The purpose of the invention is realized as follows: a pain recognition system based on confidence interval fusion threshold criterion comprises a monitoring data acquisition module, a threshold processing module, a pain recognition module, an output module and an alarm module;

the monitoring data acquisition module is used for acquiring multidimensional monitoring data, acquiring vital sign data of a patient through the vital sign detector, acquiring sleep condition data of the patient through the sleep induction device, and inputting pain grading monitoring data of the patient through the medical care information system;

the threshold processing module is used for fusing threshold criteria to preprocess pain monitoring data and carrying out two-step threshold preprocessing on the collected vital sign data, sleep condition data and pain grading monitoring data;

the pain recognition module performs function training on the monitoring data by using the constructed pain monitoring data grading model to obtain a corresponding pain level, determines a confidence interval of pain detection data and obtains a pain monitoring value range of the maximum probability corresponding to a certain pain level;

the output module is used for displaying the pain level range corresponding to each real-time monitoring data training and transmitting the pain level range to the medical staff computer;

the alarm module is used for judging whether the monitoring data are in an effective range or not, and if the monitoring data are invalid, alarm information is sent out to prompt medical staff to process the monitoring data.

By adopting the technical scheme, compared with the prior art, the invention has the beneficial effects that: 1) the threshold processing module is fused with threshold criteria to carry out two-step preprocessing on the initial data, invalid data are removed, the efficiency of model training is improved, and the training result can be more accurately close to the true pain value of a patient; 2) the pain recognition module adopts the established neural network pain monitoring data grading model to realize real-time, accurate and comprehensive pain assessment on a pain patient, and performs corresponding real-time feedback and protection alarm according to an assessment result, so that the workload of medical workers can be reduced, more comprehensive medical guarantee and timely rescue can be provided for the patient, and the real-time monitoring on the pain symptom of the patient can be accurately, efficiently and automatically realized.

Further, the vital sign data includes heart rate value, blood pressure value, respiration value, blood oxygen saturation and body temperature of the patient; the sleep condition data includes total sleep time, sleep break time, and deep sleep ratio of the patient.

In order to remove invalid data and improve the efficiency of model training, so that the training result can be more accurately close to the true pain value of the patient, the two-step threshold preprocessing specifically comprises:

the method comprises the following steps of firstly, preprocessing, namely comparing all collected monitoring data with a preset threshold value, wherein the range of the preset threshold value is an area between a maximum monitoring threshold value and a minimum monitoring threshold value, and if the monitoring data are within the range of the preset threshold value, namely the numerical value of the monitoring data are within an effective range, further inputting the monitoring data into a constructed pain monitoring data grading model; if the monitoring data at the current moment is not in the range of the normal rest state or the set pain threshold value of the patient, the alarm module sends alarm information to remind medical staff of carrying out corresponding treatment;

and secondly, preprocessing, namely processing the collected vital sign data and the sleep condition data, calculating the variation of the vital sign data and the sleep condition data at a certain moment, comparing the variation with the maximum variation of the monitoring data in the rest state, and if the variation of the monitoring data at a certain moment is larger than the maximum variation in the rest state, indicating that the fluctuation of the vital sign and the sleep condition of the patient is large, further inputting the monitoring data at the moment into the constructed pain monitoring data grading model.

In order to realize real-time, accurate and comprehensive pain assessment for pain patients, the constructed pain monitoring data grading model specifically comprises a data input layer, I hidden layers and an output processing layer group.

Further, the data input layer is used for inputting real-time data corresponding to the pain monitoring index, including real-time values of vital sign data, sleep condition data and pain scores, and real-time variation of the vital sign data and real-time variation of the sleep condition data.

Furthermore, the hidden layer group is set to be I hidden layers, any ith hidden layer group sequentially comprises two convolution layers, a pooling layer and a regularization layer, and the pain level of the ith layer is represented as H_iPain was monitored in the MATLAB program for values 0 to X_tPain level H corresponding to the range_iFitting the parameter function to obtain pain level H under 5 th order limitation_i(X_t) The corresponding best restriction function:

wherein H_i(X_t) Defining an optimal defining function for the pain level at 5 th order; x is the number of_tReal-time specific values for a certain pain monitoring index; alpha is alpha₁，α₂，α₃，α₄，α₅，α₆Best fit coefficients defining a function for order 5; setting the initial dimension of the hidden layer as 100, and then setting the dimension of the hidden layer according to the obtained H_iAnd true experimental value H₀The dimension of the next hidden layer is calculated according to the difference between the first hidden layer and the second hidden layer, and the dimension of the ith hidden layer is expressed as:

further, the output processing layer group comprises a planarization layer, a first full-connection layer, an I +1 th regularization layer and a second full-connection layer; the second full-connection layer connects the input feature vectors of the I +1 th regularization layer to the M pain levels of the second full-connection layer, processes the input feature vectors, obtains the feature vectors with one dimension being M, and inputs the feature vectors into the softmax regression classifier, so that the probability of pain monitoring data corresponding to the M pain levels is output, and training is completed.

To further improve the accuracy of the model prediction, said determining the confidence interval of the pain detection data comprises in particular inputting a vital sign value X_tAnd corresponding pain level H_iEquation f (X) introduced into the MATLAB program under the conditions of the Weibull equation_t) Automatic fitting is carried out, and the Weibull equation mode is obtained as follows:

in the formula, f (X)_t) Monitoring the value X for pain_tExpressed as the number of occurrences or probability of each pain monitoring value in a certain pain level; b₁Is the shape parameter of Weibull equation; b₂Dimension parameters of Weibull equation; b₃Position parameters of Weibull equation; e is the number of fingers; t is numerical numbering of pain monitoring data, f (X) is taken by drawing Weibull probability distribution map_t) The 95% confidence interval is the pain monitoring range of values that represents the maximum probability for a certain pain level.

Drawings

FIG. 1 is a block diagram of the system of the present invention.

FIG. 2 is a flow chart of the threshold module of the present invention for preprocessing pain monitoring data.

FIG. 3 is a schematic diagram of a pain monitoring data grading model according to the present invention.

Fig. 4 is a diagram of a pain level data analysis of pain monitor data values according to the present invention (heart rate values are used as an example).

FIG. 5 is a diagram of a pain level data analysis of the variation of pain monitoring data according to the present invention (for example, the variation of systolic blood pressure).

Detailed Description

Fig. 1 shows a pain recognition system based on a confidence interval fusion threshold criterion, which includes a monitoring data acquisition module, a threshold processing module, a pain recognition module, an output module, and an alarm module;

the monitoring data acquisition module is used for acquiring multidimensional monitoring data, acquiring vital sign data of a patient through the vital sign detector, acquiring sleep condition data of the patient through the sleep induction device, and inputting pain grading monitoring data of the patient through the medical care information system; vital sign data includes heart rate values, blood pressure values, respiration values, blood oxygen saturation, and body temperature of the patient; the sleep condition data comprises total sleep time, sleep interruption time and deep sleep ratio of the patient; a patient numerical score (NRS) entered by the healthcare information system as a numerical value ranging from 0 to 10, with 0 being no pain and 10 being intolerable pain.

As shown in fig. 2, the threshold processing module is configured to perform pre-processing on pain monitoring data by fusing a threshold criterion, and perform two-step threshold pre-processing on the collected vital sign data, sleep condition data, and pain score monitoring data;

the method comprises the steps of firstly, preprocessing, namely comparing all collected monitoring data with a preset threshold, wherein the range of the preset threshold is an area between a maximum monitoring threshold X2 and a minimum monitoring threshold X1, and if the monitoring data are within the range of the preset threshold, namely the numerical value of the monitoring data are within an effective range, further inputting the monitoring data into a constructed pain monitoring data grading model; if the monitoring data at the current moment is not in the range of the normal rest state or the set pain threshold value of the patient, the alarm module sends alarm information; the alarm module is used for judging whether the monitoring data is in an effective range or not, and if the monitoring data is invalid, the alarm module sends alarm information to prompt medical staff to process the monitoring data.

The specific method of the first step of pretreatment comprises the following steps: the heart rate value monitored by the vital sign monitor is recorded as A, the blood pressure value comprises systolic pressure B and diastolic pressure B, the respiratory value is recorded as C, the blood oxygen saturation is recorded as D, and the body temperature is recorded as E; the total sleep time monitored by the sleep sensor is recorded as F, the sleep time G, the sleep interruption time H, the deep sleep ratio I, and the pain score recorded by the medical care information system is recorded as J.

When A1 ≦ A2, indicating that the patient's heart rate is within the valid range, where A1 and A2 represent the minimum and maximum threshold values, respectively, for the patient's heart rate at pain;

when B1 is less than or equal to B2, the systolic pressure of the patient is in an effective range, wherein B1 and B2 respectively represent the minimum and maximum threshold values of the systolic pressure when the patient is painful;

when b1 is less than or equal to b2, the diastolic pressure of the patient is in a valid range, wherein b1 and b2 respectively represent the minimum and maximum threshold values of the diastolic pressure of the patient when the patient is painful;

by analogy … …

When I1 is less than or equal to I2, the deep sleep ratio of the patient is in an effective range, wherein I1 and I2 respectively represent the minimum and maximum threshold values of the deep sleep ratio when the patient is painful;

a pain score for the patient is indicated to be within a valid range when J1 ≦ J2, where J1 and J2 indicate minimum and maximum thresholds, respectively, for the pain score for the patient.

Since the vital sign and sleep condition monitoring data indirectly reflect the pain condition of the patient and are greatly influenced by individual differences, the collected vital sign and sleep condition monitoring data enter the second step of processing;

the second step of preprocessing, namely processing the collected vital sign data and the sleep condition data, calculating the variation of the vital sign data and the sleep condition data at a certain moment, and comparing the variation with the maximum variation of the monitoring data in a rest stateComparing, if the variation of the monitoring data at a certain momentGreater than the maximum variation in the resting stateAnd if the vital signs and the sleep condition of the patient fluctuate greatly, the monitoring data at the moment are further input into the constructed pain monitoring data grading model.

The specific method of the second step of treatment is as follows: data of a patient in a resting state measured by a vital sign monitor: heart rate data set is a_{Combination of Chinese herbs}＝{a₁,a₂,a₃,…,a_nThe systolic pressure data set is B_{Combination of Chinese herbs}＝{b_{Height 1},b_{Height 2},b_{Height 3},…,b_{High n}The diastolic pressure data set is b_{Combination of Chinese herbs}＝{b_{Low 1},b_{Low 2},b_{Low 3},…,b_{Low n is}The respiratory data set is C_{Combination of Chinese herbs}＝{c₁,c₂,c₃,…,c_nThe blood oxygen saturation data set is D_{Combination of Chinese herbs}＝{d₁,d₂,d₃,…,d_nThe temperature data set is E_{Combination of Chinese herbs}＝{e₁,e₂,e₃,…,e_n}; measuring data of the patient in a resting state by the sleep sensing device: total sleep time dataset is F_{Combination of Chinese herbs}＝{f₁,f₂,f₃,…,f_nThe data set of sleep time is G_{Combination of Chinese herbs}＝{g₁,g₂,g₃,…,g_nH is the sleep interruption time data set_{Combination of Chinese herbs}＝{h₁,h₂,h₃,…,h_nThe deep sleep proportion data set is I_{Combination of Chinese herbs}＝{i₁,i₂,i₃,…,i_n}. And respectively calculating the vital sign data of the patient at a specific moment and the variation of the sleep state data of a certain day according to a formula by taking the monitoring value of the patient in the rest state as a reference.

When | a_t-a_min|＞|a_max-a_minWhen the variation of the heart rate data of the patient at a certain moment is larger than the maximum variation in the rest state, entering a pain grading training model, wherein a_tRepresenting the heart rate at a certain moment in time in the patient monitoring state,a_maxand a_minRepresenting the maximum and minimum heart rate of the patient at rest;

when | b_{High t}-b_{High min}|＞|b_{High max}-b_{High min}When the variation of the systolic pressure data at a certain moment of the patient is larger than the maximum variation in the resting state, | the pain classification training model is entered, wherein b_{High t}Representing the systolic pressure at a certain moment in time under the patient's monitoring regime, b_{High max}And b_{High min}Representing the maximum and minimum systolic pressures at rest for the patient;

when | b_{Low t}-b_{Low min}|＞|b_{Low max}-b_{Low min}When the variation of the diastolic pressure data of the patient at a certain moment is larger than the maximum variation in the resting state, entering a neural network model, wherein b_{Low t}Indicating the diastolic pressure at a certain moment in the monitored state of the patient, b_{Low max}And b_{Low min}Representing the maximum and minimum diastolic pressures at rest in the patient;

by analogy … …

When | h_t-h_min|＞|h_max-h_minIf the variation of the sleep interruption time data of a certain day of the patient is larger than the maximum variation in the resting state, entering a pain grading training model, wherein h_tIndicating the sleep interruption time of a certain day, h, in a patient monitoring state_maxAnd h_minRepresenting maximum and minimum sleep interruption times for a patient at rest;

when | i_t–i_max|＞|i_min-i_maxWhen the variation of the deep sleep ratio data of a certain day of the patient is larger than the maximum variation in the rest state, entering a pain grading training model, wherein i_tIndicating the deep sleep fraction of a certain day under patient monitoring i_maxAnd i_minRepresenting the maximum and minimum deep sleep fraction of the patient at rest.

The pain recognition module performs function training on the monitoring data by using the constructed pain monitoring data grading model to obtain a corresponding pain level, determines a confidence interval of pain detection data and obtains a pain monitoring numerical range of the maximum probability corresponding to a certain pain level;

as shown in fig. 3, the constructed pain monitoring data grading model specifically includes a data input layer, I hidden layers and an output processing layer group;

the data input layer is used for inputting real-time data corresponding to the pain monitoring indexes, and the real-time data comprises vital sign data (heart rate P, blood pressure BP, respiration R, blood oxygen saturation SpO2 and body temperature T), sleep condition data (total sleep time, sleep interruption time and deep sleep ratio) and real-time numerical values of pain scores, real-time variable quantities of the vital sign data (heart rate P, blood pressure BP, respiration R, blood oxygen saturation SpO2 and body temperature T) and real-time variable quantities of the sleep condition data (total sleep time, sleep interruption time and deep sleep ratio).

The hidden layer group is set as I hidden layers, any ith hidden layer group sequentially comprises two convolution layers, a pooling layer and a regularization layer, and the pain level of the ith layer is represented as H_iPain was monitored in the MATLAB program for values 0 to X_tPain level H corresponding to the range_iFitting the parameter function to obtain pain level H under 5 th order limitation_i(X_t) The corresponding best restriction function:

wherein H_i(X_t) Defining an optimal defining function for the pain level at 5 th order; x is the number of_tReal-time specific values for a certain pain monitoring index; alpha is alpha₁，α₂，α₃，α₄，α₅，α₆Best fit coefficients defining a function for order 5; setting the initial dimension of the hidden layer as 100, and then setting the dimension of the hidden layer according to the obtained H_iAnd true experimental value H₀The dimension of the next hidden layer is calculated according to the difference between the first hidden layer and the second hidden layer, and the dimension of the ith hidden layer is expressed as:

the specific calculation process is as follows: the pain monitoring data enter a first hidden layer after being processed by a threshold value, the dimensionality of the first hidden layer is 100 dimensionalities, and any n pain monitoring data obtain corresponding pain levels H after being processed by the first hidden layer₁Thus the first pain monitoring data corresponds toWherein x is_tInputting a specific numerical value at a certain moment of a certain monitoring numerical value into a second hidden layer after function operation and first Relu nonlinear mapping;

the dimension of the second hidden layer isThe pain value calculated by the first layer function is obtained by processing the second hidden layerWherein x is_tInputting a specific value at a certain moment of a certain monitoring value into a third hidden layer after the specific value is subjected to second Relu nonlinear mapping;

……

the dimension of the ith hidden layer isThe pain value calculated by the function of the previous layer is obtained by processing the ith hidden layerAnd the data is input into an output layer after the ith Relu nonlinear mapping.

The output processing layer group comprises a planarization layer, a first full-connection layer, an I +1 th regularization layer and a second full-connection layer; the second full-connection layer connects the input feature vectors of the I +1 th regularization layer to the M pain levels of the second full-connection layer, processes the feature vectors, obtains the feature vectors with one dimension being M, and inputs the feature vectors into the softmax regression classifier, so that the probability of pain monitoring data corresponding to the M pain levels is output, and training is completed.

To further improve the accuracy of the model prediction, the determination of the confidence interval of the pain detection data comprises in particular the input of a vital sign value X_tAnd corresponding pain level H_iEquation f (X) introduced into the MATLAB program under the conditions of the Weibull equation_t) Automatic fitting is carried out, and the Weibull equation mode is obtained as follows:

in the formula, f (X)_t) Monitoring the value X for pain_tExpressed as the number of occurrences or probability of each pain monitoring value in a certain pain level; b₁Is the shape parameter of Weibull equation; b₂Dimension parameters of Weibull equation; b₃Position parameters of Weibull equation; e is the number of fingers; t is the numerical number of the pain monitoring data.

As shown in FIGS. 4 and 5, f (X) is taken by plotting Weibull probability distribution map_t) The 95% confidence interval is the pain monitoring range of values that represents the maximum probability for a certain pain level.

The output module is used for displaying the pain level range corresponding to each real-time monitoring data training and transmitting the pain level range to the medical staff computer; and finally, the medical staff finally determines the pain condition of the patient according to the weight occupied by the data according to the pain monitoring numerical range corresponding to each pain level obtained by the pain monitoring data grading model.

The present invention is not limited to the above-mentioned embodiments, and based on the technical solutions disclosed in the present invention, those skilled in the art can make some substitutions and modifications to some technical features without creative efforts according to the disclosed technical contents, and these substitutions and modifications are all within the protection scope of the present invention.