WO2023070701A1

WO2023070701A1 - Automatic animal pain test system

Info

Publication number: WO2023070701A1
Application number: PCT/CN2021/128364
Authority: WO
Inventors: 谢曦; 钟成锦; 王浩; 李柏鸣; 李仁杰; 陈惠琄
Original assignee: 中山大学
Priority date: 2021-10-29
Filing date: 2021-11-03
Publication date: 2023-05-04
Also published as: CN114010155B; CN114010155A

Abstract

An automatic animal pain test system, comprising a moving apparatus, a camera apparatus, and a computer device (8) connected to the camera apparatus. The moving apparatus comprises a moving box body (2), a microneedle array (5) located in the moving box body (2), and an angle adjustment support (4). The camera apparatus is configured to photograph a motion track video and send same to the computer device (8). The computer device (8) comprises at least one processor to implement the following steps: obtaining the motion track video, and inputting the motion track video into a trained recognizer to obtain position information; determining coordinate information according to the position information, and processing the coordinate information to obtain motion feature information; and inputting the motion feature information into a trained prediction model to obtain a predicted pain value. Animal pain can be automatically tested, time and labor are saved, the accuracy is high, and the system can be widely applied to the field of medical instruments.

Description

An automated animal pain testing system

technical field

The invention relates to the field of medical equipment, in particular to an automatic animal pain testing system.

Background technique

Pain has a major impact on human health because it is very difficult to treat and is widely associated with various diseases. Understanding the cause of pain and quantifying the degree of pain has been a key challenge in medical history and modern medicine. The translation of basic research on pain into clinical applications is limited by the lack of effective tools to quantify pain. Pain assessment is often subjective and limited to direct human use due to ethical concerns. Rodents such as mice and rats are often used as standard models to measure their responses to noxious stimuli. Pain testing methods have greatly enhanced our understanding of the genes, neurotransmitters, and cell receptors involved in the mechanisms of pain perception, as well as the discovery of pain therapeutic drugs. However, the traditional manual measurement method is time-consuming and laborious, and the measurement results are easily affected by the outside world.

Contents of the invention

In view of this, the purpose of the embodiments of the present invention is to provide an automated animal pain testing system, which can automatically test animal pain, saves time and effort, and has high accuracy.

The embodiment of the present invention provides a kind of automatic animal pain test system, comprises activity device, camera device and the computer equipment connected with described camera device; Wherein,

The movable device includes a movable box, an array of microneedles located in the movable box, and an angle adjustment bracket for adjusting the movable box;

The camera device is used to take a video of the movement track of the animal in the movable box, and send the video of the movement track to the computer device;

Said said computer equipment includes:

at least one processor;

at least one memory for storing at least one program;

When the at least one program is executed by the at least one processor, the at least one processor is made to implement the following steps:

Obtaining the motion trajectory video, and inputting the motion trajectory video into a trained recognizer to obtain position information;

determining coordinate information according to the position information, and processing the coordinate information to obtain motion feature information; the coordinate information includes coordinate values and time series;

Input the motion feature information into the trained prediction model to obtain the predicted pain value.

Optionally, the training method of the recognizer includes:

Obtaining a motion track video sample, and determining the animal position in the motion track video sample according to the frame difference method and the region of interest;

Collecting positive sample pictures and negative sample pictures in the motion trajectory video samples according to the animal position;

The positive sample picture and the negative sample picture are input to the haar cascade classification model for training;

The haar cascade classification model whose recognition accuracy reaches the preset accuracy is used as the recognizer.

Optionally, the determining coordinate information according to the location information specifically includes:

Determine the outline of the active box according to the carry edge detection algorithm and the Hough transform line detection algorithm;

The coordinate system and the coordinate information of the animal in the motion track video are determined according to the position information and the outline.

Optionally, the processor also implements the following steps:

Resample the coordinate information by using a linear difference method to obtain a first coordinate value and a first time series including a preset number of frames within a preset time, and combine the first coordinate value and the first time series as coordinate information.

Optionally, the processor also implements the following steps:

performing center smoothing filtering on the first coordinate value and the first time series using a sliding window to obtain a second coordinate value and a second time series, and using the second coordinate value and the second time series as coordinate information .

Optionally, the processing the coordinate information to obtain motion feature information specifically includes:

Deriving the coordinate information to obtain a velocity time series;

determining a velocity position distribution histogram according to the velocity time series and the coordinate information;

Integrating the speed time series to obtain a distance time series;

A histogram of distance location distribution is determined according to the distance time series and the coordinate information.

Optionally, the training method of the prediction model includes:

Obtain the dynamic characteristic information and the corresponding pain value of the animal test sample;

converting the kinetic feature information of the animal test sample into a two-dimensional matrix, and standardizing the two-dimensional matrix;

The principal components are extracted from the standardized two-dimensional matrix, and the two principal components with the best correlation are binary fitted to obtain a fitted straight line between the pain value and the motion feature, and the fitted straight line is used as a prediction model.

Optionally, the training method of the prediction model includes:

According to the standardized two-dimensional matrix and the corresponding pain value, the classification neural network is trained, and the classification neural network with prediction accuracy is used as the prediction model; the classification neural network includes one layer of input layer, three layers of hidden layers and one layer of output layer.

Optionally, the movable box includes a base and a baffle, and the base and the baffle are assembled through a male and female tenon. The movable box is divided into a rest area and an activity area. The rest area and the The active area is isolated by a baffle, and the baffle is assembled with the base by means of a male and a female tenon.

Optionally, the entire array of microneedles is installed in the active area, and the radius of the top of the entire array of microneedles gradually decreases from the end near the rest area.

Implementing the embodiment of the present invention includes the following beneficial effects: the embodiment of the present invention generates pain stimulation for the animal to be tested through the mobile device; the motion track video of the animal to be tested is captured by the camera device; the position information and coordinate information are extracted from the motion track video by computer equipment and motion feature information, and transfer the motion feature information to the trained prediction model to obtain the predicted pain value; thereby completing the automatic test of animal pain, the measurement does not require manual intervention, saves time and effort, and has high accuracy.

Description of drawings

Fig. 1 is the structural representation of a kind of automated animal pain test system provided by the embodiment of the present invention;

FIG. 2 is a schematic flowchart of steps performed by a processor according to an embodiment of the present invention;

Fig. 3 is an interface diagram showing animal coordinate information provided by an embodiment of the present invention;

Fig. 4 is an interface diagram for displaying animal movement characteristic information provided by an embodiment of the present invention.

Detailed ways

The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. For the step numbers in the following embodiments, it is only set for the convenience of illustration and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art sexual adjustment.

Referring to Fig. 1 and shown in Fig. 2, the embodiment of the present invention provides a kind of automatic animal pain testing system, comprises movable device, camera device and the computer equipment that is connected with described camera device; Wherein,

Said said computer equipment includes:

at least one processor;

at least one memory for storing at least one program;

S100. Obtain the motion track video, and input the motion track video to a trained recognizer to obtain position information;

S200. Determine coordinate information according to the position information, and process the coordinate information to obtain motion feature information; the coordinate information includes coordinate values and time series;

S300. Input the motion feature information into the trained prediction model to obtain the predicted pain value.

In Figure 1, black light barrier cloth 1, mouse activity box 2, infrared night vision camera 3, angle adjustment bracket 4, gradient microneedle array plate 5 in the bottom active area, friction plate 6 in the rest area, data transmission line 7, computer equipment 8. Infrared LED light source9.

Those skilled in the art can understand that when the movable device is in a horizontal state, the probability of the animal to be tested moving in the area where the microneedles are arranged is relatively small. An angle adjustment bracket to adjust the angle of the movable unit. The angle adjustment bracket makes a certain angle between the movable device and the ground, and the activity of the mouse will be affected by the gravity to move towards the whole array of microneedles.

It should be noted that the animal to be tested in this embodiment is a mouse, and the dark environment in which the mouse lives should be simulated as much as possible, and a black light-shielding barrier cloth is covered outside the movable device. The LED light group is used as the camera light source.

It should be noted that the computer equipment may be different types of electronic equipment, including but not limited to terminals such as desktop computers and laptop computers.

It should be noted that the test system supports multiple systems to work together, which can greatly save experimenters' experiment time.

In order to better observe and record the behavioral information of the experimental animals in the device, the front baffle of the experimental device is made of transparent acrylic material and cut by laser, and the rest of the device is made of acrylic material with light-shielding stickers.

Those skilled in the art can understand that the base and the baffle are assembled by a male and female tenon, and the baffle between the rest area and the activity area is assembled by a male and female tenon, which is convenient for assembly and cleaning.

Specifically, the microneedle array in this embodiment includes 4 sub-arrays, the height of the entire array of microneedles is 3 mm, the interval between microneedles is 1.8 mm, and 4 microneedle sub-arrays of 12 cm × 6 cm form a microneedle array. Needle runway. Among them, the top radius of the microneedles of the No. 1 array gradually changed from 0.03 mm to 0.08 mm, the microneedle top radius of the No. 2 array gradually changed from 0.09 mm to 0.14 mm, and the microneedle radius of the No. 3 array gradually changed from 0.15 mm to 0.20 mm, the microneedle radius of the No. 4 array was gradually changed from 0.21 mm to 0.26 mm; the No. 4 array was close to the end of the resting area. In the present embodiment, the active area of the movable box is 48cm long, 6cm wide, and 20cm high, and the rest area is 12cm long, 6cm wide, and 20cm high.

In the microneedle array in this embodiment, the sharpest microneedles with a top radius of 0.03 mm are sharp enough to induce paw withdrawal in mice, but they will not puncture the skin of their feet. The farther the mouse's position is from the rest area, the sharper the top of the microneedles, the higher the pain level; when the experimental mice with a lower pain threshold enter the active area from the rest area, due to their lower pain threshold, The less pain it can tolerate, the less active it will be.

Those skilled in the art can understand that, based on the design of the gradient microneedle array, different degrees of pain can be produced on the feet of mice. Due to the pain effect, mice tend to avoid the sharp microneedle area and prefer to move in the non-sharp microneedle area. By designing microneedles with continuous gradient changes in diameter, the degree of pain caused to mice also has continuous changes. The gradient microneedle array can stably form specific pain to mice, and these pain behavior characteristics can be significantly different from other non-pain behavior characteristics of mice, and are easier to be extracted and analyzed.

Experiments have proved that gradient microneedles can produce different pain levels and distinguish different pain states so as to produce different pain characteristics. When tilted at 0 degrees, non-gradient change plate, the mouse movement characteristics are very inconspicuous, and the frequencies distributed in each position are basically the same, showing no characteristic difference. The location characteristics are basically the same; when the slope is 0 degrees, the gradient plate has a significant difference in its location distribution characteristics; the area with thicker microneedles has a higher distribution frequency of mice; the area with sharper microneedles, the distribution frequency of mice The distribution frequency of is low, so it can produce obvious changes in the position characteristics; selecting a certain inclination angle is helpful to promote the movement of non-moving mice. Therefore, the non-gradient change plate, no matter how much the pain threshold of the mouse is (the mouse is afraid of pain or not pain), all the microneedles have the same sharpness, and the extracted motion features are not much different; the gradient change plate is helpful to produce significantly changed features, in order to relate these features to pain values.

The experimental process is as follows: First, the activity device should be built according to the design drawings, the mouse should be placed in a dark environment, and the infrared LED light source on the front baffle of the device should be turned on, and the camera on the front of the experimental device should be turned on to test whether it can work normally. Then, place the mice in the rest area of the activity device to rest for 5 minutes. After the mice adapt to the environment of the experimental device, turn on the camera and open the barrier between the rest area and the experimental area. After the area moves to the experimental area, close the barrier; finally, when the experimental mouse has been active in the experimental area for 15 minutes, turn off the camera and transfer the mouse from the experimental device to the mouse cage.

Optionally, the training method of the recognizer includes:

S110. Obtain a motion track video sample, and determine the position of the animal in the motion track video sample according to the frame difference method and the region of interest;

S120. Collect a positive sample picture and a negative sample picture in the motion track video sample according to the animal position;

S130. Input the positive sample picture and the negative sample picture into the haar cascade classification model for training;

S140. Using a haar cascade classification model whose recognition accuracy reaches a preset accuracy rate as a recognizer.

It should be noted that the haar cascade classification model includes AdaBoost algorithm, cascade, Haar-like features, fast calculation of integral graph, etc.

Specifically, first use the frame difference method to roughly locate the position of the moving mouse in each frame of the video, perform a binarization threshold on each frame of the image and use connected domain analysis to find the approximate position of the mouse; then set the ROI The area excludes irrelevant interference. The position of the mouse is determined according to the frame difference method and the ROI area, and a large number of positive sample pictures are collected according to the center of the position, and the remaining pictures are used as negative samples. The collection of negative sample pictures is opposite to the collection of positive sample pictures, and the method of random position and random size interception is adopted. Put the positive and negative samples collected in batches into the haar cascade classification model for training, and train a preliminary version of the recognizer with a recognition accuracy of about 80%. Using the recognizer to locate the mouse position more accurately, when the recognizer appears When the positive sample image cannot be detected, use the frame difference method and the ROI area to supplement the undetected mouse posture, and then put these mouse postures that cannot be detected by the haar recognizer into the haar cascade classifier model again. Training, multiple iterations Stop iterations when the recognition accuracy reaches 99.9%, and get the final version of the haar cascade mouse recognizer. Use the final version of the haar cascade mouse recognizer to identify and locate the mouse position information for each frame in the video.

It should be noted that, in order to save the sample collection time of the training recognizer, an automatic sample collector can also be designed to automatically supplement positive samples, negative samples, and some undetected mouse gestures to enhance the recognition rate of the recognition algorithm and Accuracy.

S210. Determine the contour of the movable box according to the carry edge detection algorithm and the Hough transform line detection algorithm;

S220. Determine a coordinate system and coordinate information of the animal in the motion track video according to the position information and the outline.

Specifically, adaptive histogram equalization is used to optimize edge information by enhancing contrast in the form of a sliding window. Use the canny edge detection algorithm to detect the edge in the image, use the statistical Hough transform to detect the result of the canny edge in the picture, and obtain the straight line segment in the edge; then perform cluster analysis on the obtained straight line segment to obtain the glass The clustered line segments near the edge of the box are collinearly judged to recombine the collinear line segments into a straight line; the longest straight line is found, and the longest straight line in the cluster is obtained by analogy, which is the edge of the area, and then according to the cuboid edge The pairwise parallel rule optimizes for instabilities in the video to ensure the most accurate outline of the glass box.

Specifically, according to the detected active area and actual size of the mouse, the pixel position of the mouse is transformed into a world coordinate value through coordinate transformation. Save the world coordinate value of each frame, and output the position time series of the entire video to txt; and display the current lateral and vertical trajectory of the mouse in real time, as well as the distribution histogram of the mouse in each position. Coordinate transformation includes two methods: fully automatic calibration coordinate transformation and manual calibration coordinate transformation. In some harsh scenes, the results of fully automatic calibration may not be particularly good. Manual calibration can be selected, and the calibration map of each video will be saved for use. Observe the calibration plot where the problem occurs. As shown in Figure 3, Figure 3 is an interface displaying coordinate information during the experiment.

Optionally, the processor also implements the following steps:

S230. Resample the coordinate information by using the linear difference method to obtain a first coordinate value and a first time series including a preset number of frames within a preset time, and combine the first coordinate value and the first Time series as coordinate information.

Optionally, the processor also implements the following steps:

S240. Perform center smoothing filtering on the first coordinate value and the first time series using a sliding window to obtain a second coordinate value and a second time series, and use the second coordinate value and the second time series as coordinate information.

Specifically, the obtained time series of mouse positions were resampled using the linear difference method, and the original 25-35 frames per second (camera frame rate ranging from 25 frames to 35 frames) were uniformly resampled to 60 frames per second. frame. In view of the shaking phenomenon of mice, which affects the calculation of subsequent speed, acceleration and other parameters, a 1-second sliding window is used to perform center smoothing filtering on the data.

S250. Deriving the coordinate information to obtain a velocity time series;

S260. Determine a velocity position distribution histogram according to the velocity time series and the coordinate information;

S270. Integrate the speed time series to obtain a distance time series;

S280. Determine a distance location distribution histogram according to the distance time series and the coordinate information.

For the resampled and filtered position time series, derive the velocity time series, combine the velocity time series with the position time series to obtain the velocity position distribution histogram, integrate the velocity time series to obtain the distance time series, and combine the distance time series with the position time series A histogram of the distance distribution can be obtained. Display the obtained position distribution histogram, speed position distribution histogram, and distance position distribution histogram on the animal behavior data processing and analysis software for real-time observation. As described in Figure 4, Figure 4 is an interface displaying motion feature information during the experiment.

According to the above calculation process, 12 features of position frequency distribution, 20 features of velocity frequency distribution, 12 features of position distance frequency distribution, 12 features of position distance distribution, 1 feature of total distance, 12 features of position positive velocity distribution, 12 characteristics of position negative velocity distribution, 12 characteristics of acceleration frequency distribution, 12 characteristics of position positive acceleration distribution, 12 characteristics of position negative acceleration distribution, 15 characteristics of longitudinal position frequency distribution, 14 characteristics of longitudinal velocity frequency distribution, longitudinal position distance 12 features are distributed, for a total of 166 features. The video sequence is intercepted for 300s, 600s, and 900s, each time sequence has 166 features, and the total number of features: 166*3=498.

Optionally, the training method of the prediction model includes:

S310A. Acquiring the dynamic characteristic information of the animal test sample and the corresponding pain value;

S320A. Convert the motion characteristic information of the animal test sample into a two-dimensional matrix, and standardize the two-dimensional matrix;

S330A. Extract principal components from the standardized two-dimensional matrix, and perform binary fitting on the two principal components with the best correlations to obtain a fitted straight line between the pain value and the motion feature, and use the fitted straight line as a prediction model.

Specifically, it is first necessary to establish a fitting prediction model training library. For example, if the number of mice in the training library is 300, each mouse can calculate 498 features to form 300 pieces of 498-dimensional data, which is a 300*498 binary model. Dimensional motion feature matrix, the output 300*498 features are standardized by column using the z-score algorithm to form a matrix X, and the 498 values in each row of the matrix X are zero-meaned, that is, the mean value of this row is subtracted to obtain New matrix, and then find the covariance matrix C of the matrix, and then find the eigenvalues and corresponding eigenvectors of the covariance matrix; then arrange the eigenvectors into a matrix from top to bottom according to the size of the corresponding eigenvalues, and take The first two lines constitute the weight matrix P, and Y=PX is the result matrix after the principal component is proposed, the size is 300*2, and Y represents the matrix corresponding to the pain value of the mouse. Two principal components with better correlation among the extracted principal components were subjected to binary fitting to obtain a fitted straight line, and the mouse pain value was predicted according to the fitted straight line.

Optionally, the training method of the prediction model includes:

S310B. Acquiring the dynamic characteristic information of the animal test sample and the corresponding pain value;

S320B. Convert the motion characteristic information of the animal test sample into a two-dimensional matrix, and standardize the two-dimensional matrix;

S330B. Train the classification neural network according to the standardized two-dimensional matrix and the corresponding pain value, and use the classification neural network with prediction accuracy as a prediction model; the classification neural network includes an input layer, three hidden layers and a layer output layer.

Specifically, it is first necessary to establish a fitting prediction model training library. The number of mice in the training library is selected to be 300, and each mouse can calculate 498 features to form 300 pieces of 498-dimensional data, which is a 300*498 two-dimensional In the motion feature matrix, the output 300*498 features are standardized by column using the z-score algorithm to obtain standardized mouse features, and the extracted mouse feature information and pain threshold are input to the classification neural network for training. The artificial intelligence network uses a fully connected classification network composed of multi-layer perceptrons, which consists of an input layer, three hidden layers and an output layer. Normalize the characteristic information of the mice according to the corresponding top data, divide the secondary variable data of each interval by the top secondary variable data under the corresponding variable, and eliminate the influence of individual mice due to their own activity ability and endurance, which is more It is beneficial to highlight the characteristics of pain caused by microneedles; secondly, we combine the pressure caused by the sharpness of the microneedles in each interval of the gradient microneedle array into the behavioral characteristics of mice in each interval; the normalization of each interval After normalization, the value of the secondary variable is divided by the microneedle sharpness of the corresponding interval, so that the influence of the sharpness of the microneedle in each interval on the behavior of the mouse is considered; all variables after normalization and sharpness transformation, and The general characteristics of mice such as temperature, humidity, total distance, and weight are also considered as auxiliary variables, and a total of 498 variables are used as the final input layer data. The hidden layer consists of three linear layers and two tanh activation functions. The data generated by the input layer is used as the input of the hidden layer. The input size of the hidden layer during training is 498, which is the number of features. Layer and two activation functions, the output size is 10 classification results (ten pain levels). The output layer receives the output result of the hidden layer, and calculates the result of size 10 through the Softmax function to obtain the predicted probability of ten pain values, so the result of the output layer is the classification probability of size 10. The loss function of this model adopts the Cross Entropy cross entropy loss function, and the number of training is set to 20,000 times. According to the predicted loss, after 20,000 updates, the corresponding coefficients w and b are continuously updated, and finally a training model is obtained.

It should be noted that, in order to establish a training library for artificial intelligence algorithms, different degrees of pain can be produced by injecting CFA (complete Freund's adjuvant) solution or lidocaine solution, or applying lidocaine solution locally on the paw. of mice. The CFA solution induced inflammatory pain in the mice's paws, which, depending on the dose injected, produced a group of mice with lower pain thresholds. Conversely, injection or topical application of lidocaine would likely alleviate localized nociceptive pain, producing groups of mice with higher pain thresholds. These mice were randomly mixed with untreated mice, and their pain sensitivity was tested by the fibrillar analgesia method, and then tested by the gradient microneedle array system to collect movement curves. In order to build a larger AI training sample library, covering highly varying pain levels and motor characteristics, these mice were repeatedly tested to collect data for 3-5 weeks, which minimized the number of mice used.

The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the described embodiments, and those skilled in the art can also make various equivalent deformations or replacements without violating the spirit of the present invention. , these equivalent modifications or replacements are all within the scope defined by the claims of the present application.

Claims

An automated animal pain testing system is characterized in that it includes a mobile device, a camera and computer equipment connected to the camera; wherein,

The movable device includes a movable box, an array of microneedles located in the movable box, and an angle adjustment bracket for adjusting the movable box;

The camera device is used to take a video of the movement track of the animal in the movable box, and send the video of the movement track to the computer device;

Said said computer equipment includes:

at least one processor;

at least one memory for storing at least one program;

When the at least one program is executed by the at least one processor, the at least one processor is made to implement the following steps:

Obtaining the motion trajectory video, and inputting the motion trajectory video into a trained recognizer to obtain position information;

determining coordinate information according to the position information, and processing the coordinate information to obtain motion feature information; the coordinate information includes coordinate values and time series;

Input the motion feature information into the trained prediction model to obtain the predicted pain value.
The system according to claim 1, wherein the training method of the recognizer comprises:

Obtaining a motion track video sample, and determining the animal position in the motion track video sample according to the frame difference method and the region of interest;

Collecting positive sample pictures and negative sample pictures in the motion track video samples according to the animal position;

The positive sample picture and the negative sample picture are input to the haar cascade classification model for training;

The haar cascade classification model whose recognition accuracy reaches the preset accuracy is used as the recognizer.
The system according to claim 1, wherein said determining coordinate information according to said location information specifically comprises:

Determine the outline of the active box according to the carry edge detection algorithm and the Hough transform line detection algorithm;

The coordinate system and the coordinate information of the animal in the motion track video are determined according to the position information and the outline.
The system according to claim 3, wherein the processor further implements the following steps:

Resample the coordinate information by using a linear difference method to obtain a first coordinate value and a first time series including a preset number of frames within a preset time, and combine the first coordinate value and the first time series as coordinate information.
The system according to claim 4, wherein the processor further implements the following steps:

performing center smoothing filtering on the first coordinate value and the first time series using a sliding window to obtain a second coordinate value and a second time series, and using the second coordinate value and the second time series as coordinate information .
The system according to claim 1, wherein said processing said coordinate information to obtain motion feature information specifically includes:

Deriving the coordinate information to obtain a velocity time series;

determining a velocity position distribution histogram according to the velocity time series and the coordinate information;

Integrating the speed time series to obtain a distance time series;

A histogram of distance location distribution is determined according to the distance time series and the coordinate information.
The system according to claim 1, wherein the training method of the prediction model comprises:

Obtain the dynamic characteristic information and the corresponding pain value of the animal test sample;

converting the kinetic feature information of the animal test sample into a two-dimensional matrix, and standardizing the two-dimensional matrix;

The principal components are extracted from the standardized two-dimensional matrix, and the two principal components with the best correlation are binary fitted to obtain a fitted straight line between the pain value and the motion feature, and the fitted straight line is used as a prediction model.
The system according to claim 1, wherein the training method of the prediction model comprises:

Obtain the dynamic characteristic information and the corresponding pain value of the animal test sample;

converting the kinetic feature information of the animal test sample into a two-dimensional matrix, and standardizing the two-dimensional matrix;

According to the standardized two-dimensional matrix and the corresponding pain value, the classification neural network is trained, and the classification neural network with prediction accuracy is used as the prediction model; the classification neural network includes one layer of input layer, three layers of hidden layers and one layer of output layer.
The system according to claim 1, wherein the movable box includes a base and a baffle, and the base and the baffle are assembled through a male and female tenon, and the movable box is divided into a rest area and the activity area, the rest area is separated from the activity area by a baffle, and the baffle is assembled with the base by a male and female tenon.
The system according to claim 1, wherein the array of microneedles is installed in the active area, and the radius of the top of the array of microneedles gradually decreases from the end near the rest area.