An intelligent automatic pain analysis system

Abstract

The present invention provides an intelligent pain automatic analysis system, which is characterized by including wearable devices at three locations: the hand, the wrist, and the foot, respectively used for data collection of hand touch pressure, heart rate and blood pressure, and the function of foot touch pressure changing with time, a finger EDA galvanic skin signal measuring device, which is used to connect with the wearable device on the wrist to collect hand galvanic skin signals, and a remote server. Among them, the signals generated by the wearable devices at the hand and foot locations are wirelessly transmitted to the wearable device on the wrist, and the finger galvanic skin signals measured by the wearable device on the wrist together with the galvanic skin signal measuring device are sent to the remote server. The remote server uses a pre-trained intelligent model based on the cross-signal time-domain attention mechanism to predict the pain level. It realizes the automatic analysis of the pain level using the intelligent model based on the cross-attention signal time-domain attention mechanism by the remote server on the basis of data monitoring of heart rate, blood pressure, hand and foot touch pressure, and galvanic skin, enabling the human detection of pain to be used for the study of adverse clinical drug reactions without leaving the production site outside the hospital.

Description

Technical Field

[0001] The present invention relates to an automatic pain analysis system, particularly to an intelligent automatic pain analysis system, belonging to the field of intelligent medical systems. Background Art

[0002] Existing medicine conducts research on pain by using mice in experiments and applying drugs. However, the research results from animals to humans are often not linear. Therefore, how to directly collect human pain data for studying the stimulation situation in drug clinical trials is the most accurate. Thus, an accurate drug evaluation is needed by combining the data basis of pain signals with model analysis.

[0003] On the other hand, it is also extremely important to collect data in out-of-hospital life. Thus, volunteers can conduct a large number of tests without taking time off work, thereby improving the efficiency of the tests. Therefore, how to adopt a flexible and convenient collection device is also the key to the technology.

[0004] Finally, since pain is a subjective feeling of a person and can be described in language, and the mechanism behind it is the abnormality of physiological parameters, it can be described as parameter abnormality - pain description. And the parameter abnormality is caused by factors such as drug taking. Therefore, pain can be described as a process of drug - parameter - pain description in time series, which is essentially regarded as a process of natural language generation. Therefore, if the attention mechanism is considered, it can replace humans to automatically analyze the pain level description in real time. Summary of the Invention

[0005] Based on the above problems in the prior art, the present invention will consider the following technical points: First, consider a wearable device for collecting heart rate, blood pressure, and finger and toe surface pressure data. Second, adopt an intelligent model based on the attention mechanism of the time interval between different signals to analyze pain data, so as to train the pain level based on the occurrence relationship in time among four types of different signals, namely hand, wrist, foot, and skin conductance.

[0006] Based on the above considerations, the present invention provides an intelligent automatic pain analysis system, including wearable devices at three locations: hand, wrist, and foot, which are respectively used for collecting data of hand touch pressure, heart rate and blood pressure, and foot touch pressure as functions of time, a finger EDA skin conductance signal measurement device for connecting with the wrist wearable device to collect hand skin conductance signals, and a remote server. Among them,

[0007] The signals generated by the wearable devices at the hand and foot locations are wirelessly transmitted to the wrist wearable device, and the finger skin conductance signals measured by the wrist wearable device together with the skin conductance signal measurement device are sent to the remote server, and the remote server uses a pre-trained intelligent model based on the attention mechanism of different signal time domains to predict the pain level.

[0008] Optionally, the hand wearable device includes a plurality of hand touch pressure sensors disposed on a flexible substrate, and a first integrated circuit disposed on the flexible substrate and connecting all the hand touch pressure sensors. The foot wearable device includes a flexible foot sleeve, and the flexible foot sleeve includes a bottom pad provided with a plurality of foot touch pressure sensors and a second integrated circuit provided with connections to all the foot touch pressure sensors. The wrist wearable device includes a main body and a wristband connected to the main body. The EDA skin electrical signal measuring device includes a plurality of skin electrical signal acquisition chips, which are connected to an interface detachably electrically connected to the main body through signal lines. Among them, both the first integrated circuit and the second integrated circuit include a data processor and a data wireless transmitter for communicating with the main body or directly sending data to the remote server.

[0009] The pre-training method of the intelligent model based on the heterogeneous signal time-domain attention mechanism includes the following steps:

[0010] S1 Construct the time variation functions of foot touch pressure, hand touch pressure, SCL (skin electrical signal level), and heart rate, and obtain all the peaks of SCL;

[0011] S2 For each SCL peak, divide the time span of the peak into n time elements , ,..., , and construct a long short-term memory model with each time element as a unit , and input the micro-element corresponding and the detected heart rate into the first unit, input it into the next unit through the intermediate transmission layer, and then input the detected heart rate corresponding to the micro-element into the second unit, input it into the next unit through the intermediate transmission layer, and so on, to obtain multiple output values, input them into the first function to obtain multiple outputs , forming an output matrix ; at the same time, the measured blood pressure in each time element , ,..., the formed matrix is multiplied by the output matrix to obtain the attention value , where is the heart rate matrix formed by the heart rates of n time elements;

[0012] S3 For each SCL peak, similarly follow the steps of S2 based on the same n time elements , ,..., Establish a short-term memory model of hand touch pressure , also use the second function to obtain multiple outputs and form a matrix , and form a movement value through matrix multiplication with the matrix formed by foot touch pressure , is the hand touch pressure matrix;

[0013] S4 Calculate the pain level , repeat steps S2 - S3 to obtain the pain levels of all SCL wave peaks, thus completing the construction of the model. When , it is during exercise and the pain level cannot be judged. On the contrary, when , it is emotional fluctuation, , it is mild pain, , it is moderate pain, , it is extremely painful, is the pain threshold felt by all volunteers' somatosensory evaluations;

[0014] S5 Recruit volunteers, wear wearable devices on the hand, wrist, and foot, and a finger EDA galvanic skin signal measurement device, and take drugs. The remote server collects data for training the model according to the steps of S2 - S4.

[0015] It should be understood that through the training of hand and foot touch pressure, the movement value is obtained to exclude the galvanic skin fluctuations caused by factors such as movement and emotion, and relatively pure contributions belonging to pain are filtered out.

[0016] Optionally, the hand wearable device is attached to the palm surface through a hydrogel, and the wrist wearable device is a smart watch.

[0017] Beneficial effects

[0018] Through the sensing and intelligent devices on the hand, foot, and wrist, data monitoring of heart rate, blood pressure, hand and foot touch pressure, and galvanic skin is realized, and it is handed over to the remote server for automatic analysis of the pain level by an intelligent model based on the attention mechanism in the time domain of heterogeneous attention signals. This enables the human detection of pain to be used for drug adverse clinical reaction research outside the hospital without taking off the device. Description of the drawings

[0019] Figure 1 Schematic diagram of the composition of the wearable intelligent pain automatic analysis system described in the embodiments of the present invention,

[0020] Figure 2 Function graphs of foot touch pressure, hand touch pressure, SCL, and heart rate signals changing with time,

[0021] Figure 3 ​Pre - training method algorithm logic diagram of an intelligent model based on a heterogeneous signal time - domain attention mechanism. Detailed implementation manner

[0022] As Figure 1 A wearable intelligent pain automatic analysis system is given, including a flexible substrate with hydrogel attached to the palm surface, a smartwatch including a main body with a touch screen and wristbands connected to both sides of the main body, and foot sleeves, which are respectively used for data acquisition of hand touch pressure, heart rate and blood pressure, and foot touch pressure as a function of time (as Figure 2 shown), a finger EDA galvanic skin signal measurement device, which is used to connect with a wrist - wearable device to collect hand galvanic skin signals, and a remote server.

[0023] Figure 1 On the flexible substrate with hydrogel attached to the palm surface, there are one each for the right thumb, index finger, and middle finger, and two on the palm center of the hand touch pressure sensors A. The set includes a bottom pad and a top surface. On the bottom pad, there is one heel, one front of the foot, and one big toe of the foot touch pressure sensor B. Each hand touch pressure sensor A and foot touch pressure sensor are respectively connected to the first integrated circuit at the palm heel and the second integrated circuit set on the top surface through flexible wires. Both integrated circuits have data processing and wireless transmitters, which are respectively used to process the collected hand touch pressure and foot touch pressure data and then transmit them to the remote server.

[0024] The finger EDA galvanic skin signal measurement device has three galvanic skin acquisition chips, which are set at the left thumb, index finger, and palm surface parts, and are detachably electrically connected to the interface on the host through signal wires. The host uploads the galvanic skin signals to the remote server. The remote server uses a pre - trained intelligent model based on the attention mechanism of the time interval of heterogeneous signals, and inputs the received data to predict the pain level.

[0025] Figure 2 Function graphs of foot touch pressure, hand touch pressure, SCL, and heart rate signals as a function of time are given, including two stages of before and after the subject takes medicine. For convenience, the time period from after taking medicine to the onset of pain is cut off for easy explanation. In the figure, after the onset of pain, the heart rate starts to accelerate, manifested as a shortened period, and three obvious main wave peaks (indicated by downward arrows) also appear on the SCL. The hand touch pressure function graph shows multiple peaks generated when unpacking the package when taking the medicine, and the stable pressure signal after holding the cup.

[0026] On the foot touch pressure function graph, two decreases in pressure are caused by the two lifts of the foot when leaving the ground due to pain, indicating that the experimenter moved due to pain. However, since the wave peaks of hand touch pressure and foot touch pressure are few, the factor of physical exercise can be excluded. Therefore, from the overall analysis of heterogeneous signals, to achieve intelligent automatic analysis and perception of pain, it is necessary to analyze the attention mechanism for each signal.

[0027] The pre-training method of the intelligent model based on the heterogeneous signal time-domain attention mechanism includes the following steps:

[0028] S1 Construct the time-varying functions of foot touch pressure, hand touch pressure, SCL galvanic skin signal level, and heart rate, and obtain all the peaks of SCL (for example Figure 2 );

[0029] S2 For each SCL peak, divide the time span of the peak into n time elements , ,..., , and construct a long short-term memory model with each time element as a unit , and input the micro-element corresponding and the detected heart rate into the first unit, input it into the next unit through the intermediate transmission layer, and then input the detected heart rate corresponding to the micro-element into the second unit, input it into the next unit through the intermediate transmission layer, and so on, to obtain multiple output values, input them into the first function to obtain multiple outputs , and form an output matrix ; at the same time, the blood pressure measured in each time element , ,..., the formed matrix and the output matrix are multiplied to obtain the attention value , where is the heart rate matrix formed by the heart rates of n time elements;

[0030] S3 For each SCL peak, similarly follow the steps of S2 based on the same n time elements , ,..., to establish a short-term memory model of hand touch pressure , and similarly use the second function to obtain multiple outputs , and form a matrix , and perform matrix multiplication with the matrix formed by foot touch pressure to form the motion value , is the hand touch pressure matrix;

[0031] S4 Calculate the pain level , repeat steps S2 - S3 to obtain the pain levels of all SCL peaks, and thus complete the construction of the model. When , it is during exercise and the pain level cannot be judged. On the contrary, when , it is emotional fluctuation, , it is mild pain, , it is moderate pain, , it is extremely painful, is the pain threshold felt by all volunteers' somatosensory evaluations;

[0032] Recruit volunteers, wear wearable devices on the hands, wrists, and feet, and a finger EDA skin electrical signal measurement device, take the drug, and collect data by the remote server for training the model according to the steps of S2 - S4.

Claims

1. An intelligent automatic pain analysis system, characterized in that, It includes three wearable devices, namely a hand wearable device, a wrist wearable device, and a foot wearable device, which are respectively used for data collection of hand touch pressure, heart rate and blood pressure, and the time-varying function of foot touch pressure. There is also a finger EDA galvanic skin signal measurement device, which is used to connect to the wrist wearable device to collect hand galvanic skin signals, and a remote server. Among them, the signals generated by the hand and foot wearable devices are wirelessly transmitted to the wrist wearable device, and the finger galvanic skin signals measured by the wrist wearable device together with the galvanic skin signal measurement device are sent to the remote server, and the remote server uses a pre-trained intelligent model based on the cross-signal time-domain attention mechanism to predict the pain level; the hand wearable device includes a plurality of hand touch pressure sensors arranged on a flexible substrate and a first integrated circuit arranged on the flexible substrate for connecting all hand touch pressure sensors; the foot wearable device includes a flexible foot sleeve, and the flexible foot sleeve includes a bottom pad provided with a plurality of foot touch pressure sensors and a second integrated circuit provided with connections to all foot touch pressure sensors; the wrist wearable device includes a main body and a wristband connected to the main body. The EDA galvanic skin signal measurement device includes a plurality of galvanic skin acquisition chips, which are connected to an interface detachably electrically connected to the main body through signal lines. Among them, the first integrated circuit and the second integrated circuit both include a data processor and a data wireless transmitter for communicating with the main body or directly sending data to the remote server; the hand wearable device is attached to the palm surface through hydrogel, and the wrist wearable device is a smart watch; The pre-training method of the intelligent model based on the cross-signal time-domain attention mechanism includes the following steps: S1 Construct the time-varying functions of foot touch pressure, hand touch pressure, skin conductance level (SCL), and heart rate, and obtain all the peaks of SCL; S2 For each SCL peak, divide the time span of the peak into n time elements , and construct a long short-term memory model for each time element , and input the micro-elements corresponding to and the detected heart rate into the first unit, input into the next unit through the intermediate transmission layer, and then input the detected heart rate corresponding to the micro-elements into the second unit, input into the next unit through the intermediate transmission layer, and so on, to obtain multiple output values, input into the first function to obtain multiple outputs , and form an output matrix ; at the same time, the matrix formed by the measured blood pressure in each time element and the output matrix are multiplied to obtain an attention value , where is the heart rate matrix formed by the heart rates of n time elements; For each SCL peak, similarly follow the steps of S2 based on the same n time elements Establish a hand-touch short-term memory model , and also use the second function to obtain multiple outputs , forming a matrix , and perform matrix multiplication with the matrix formed by foot touch pressure to form a motion value , , which is the hand-touch pressure matrix; S4 Calculate the pain level , repeat steps S2 - S3 to obtain the pain levels of all SCL peaks, thus completing the construction of the model. When , it is during exercise and the pain level cannot be determined. On the contrary, when , it is an emotional fluctuation, , it is mild pain, , it is moderate pain, , it is extremely painful, is the pain threshold felt by all volunteers in the somatosensory evaluation.

Images