Evaluation and training method and system based on medical infrared thermal imager

Abstract

The invention relates to the field of rehabilitation evaluation training, and relates to an evaluation and training method and system based on a medical infrared thermal imager, and the method comprises the steps: obtaining an original thermal distribution diagram; synchronously acquiring a surface electromyogram signal of the target area through an attached sensor; obtaining a denoised thermal distribution diagram; obtaining the temperature gradient of the target area according to the denoised thermal distribution diagram; judging whether the temperature gradient of the target area is greater than a temperature gradient threshold value or not and whether the amplitude of the electromyographic signal is smaller than an electromyographic signal amplitude threshold value or not; obtaining an abnormal classification result of the target area and a corresponding severity grade; and matching the corresponding training type and training intensity according to the abnormal classification result and the corresponding severity rating. By adopting the method, the accuracy of an abnormal type judgment result of the human body target area can be improved, and the efficiency of formulating a rehabilitation training scheme can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rehabilitation assessment and training. More particularly, the present application relates to an assessment and training method and system based on a medical infrared thermal imager. BACKGROUND

[0002] When a lesion occurs in a certain region of the human body, it is usually necessary to assess which type of lesion the affected area belongs to and the severity, and to develop a rehabilitation training program for the patient in order to carry out targeted rehabilitation training. In the prior art, when a certain part of the patient's body is assessed and trained, the doctor usually needs to observe the target region of the human body on site, assess the abnormal type and severity relying on his own experience, and develop a rehabilitation training program for the patient. However, this method has the risk of misjudgment when assessing, and the efficiency of developing a rehabilitation training program is relatively low. SUMMARY

[0003] To solve the technical problems of misjudgment and low efficiency of developing a training program when assessing and training a part of the human body in the prior art, the present application provides solutions in the following aspects.

[0004] In a first aspect, the present application provides an assessment and training method based on a medical infrared thermal imager, comprising:

[0005] using a medical infrared thermal imager to dynamically scan the target region and obtain an original thermal distribution map; synchronously obtaining the surface electromyogram of the target region through an attached sensor, and aligning the thermal imaging data in time and space;

[0006] using a wavelet transform algorithm to denoise the original thermal distribution map and obtain a denoised thermal distribution map;

[0007] obtaining the temperature gradient of the target region according to the denoised thermal distribution map; the temperature gradient is used to represent the temperature change rate of the target region and the surrounding tissue;

[0008] judging whether the temperature gradient of the target region is greater than a temperature gradient threshold value and whether the amplitude of the electromyogram is less than an electromyogram amplitude threshold value;

[0009] in response to the temperature gradient of the target region being greater than the temperature gradient threshold value and the amplitude of the electromyogram being less than the electromyogram amplitude threshold value, inputting the denoised thermal distribution map and the surface electromyogram into a neural network model to obtain an abnormal classification result of the target region and a corresponding severity rating;

[0010] matching a corresponding training type and training intensity according to the abnormal classification result and the corresponding severity rating.

[0011] Preferably, if the neural network model outputs multiple abnormality types, the training movements and training intensities corresponding to the abnormality types with higher priorities are selected according to the preset abnormality type priority order; the preset abnormality type priority order is inflammation>nerve damage>muscle strain.

[0012] Preferably, it also includes regional adaptive adjustment of training movements and training intensity, and the adjustment rules include: if the abnormal area is located below the patella, squatting is prohibited; if inflammation occurs in the rotator cuff muscles, the abduction angle is limited to less than or equal to 60 degrees.

[0013] Preferably, the training method of the neural network model includes:

[0014] Acquiring a data set, annotating and preprocessing the data set to form a training set, wherein the data set includes a plurality of infrared thermal images of a target area and corresponding SEMG data, wherein the plurality of infrared thermal images of the target area include infrared thermal images of a healthy target area, infrared thermal images of a target area with inflammation, infrared thermal images of a target area with muscle strain, and infrared thermal images of a target area with nerve damage;

[0015] The ResNet-18 model was adapted for the input layer and modified for the output layer to obtain a neural network model. The input layer adaptation included changing the three channels of the input layer to a single channel. The output layer modification included replacing the original 1000-class classification layer with a custom output head. The custom output head consisted of three fully connected layers and a softmax classifier. The three fully connected layers corresponded to three abnormality types: inflammation, muscle strain, and nerve damage.

[0016] Set the loss function, optimizer, hyperparameters, and regularization;

[0017] The neural network model is trained iteratively, including forward propagation, loss calculation, backpropagation, and parameter update.

[0018] Preferably, the method further includes performing short-term adaptive adjustment on the temperature gradient threshold, including:

[0019] Set the temperature gradient threshold reference value;

[0020] Obtain the patient's most recent N training data and calculate the updated temperature gradient threshold. The data of a certain training session refers to the temperature of the target area after that training session. The calculation expression for the updated temperature gradient threshold is:

[0021] T new =∝×T0+(1-∝)×median(d);

[0022] Where, T newrepresents the updated temperature gradient threshold value, a is a forgetting factor, T0 represents a temperature gradient threshold value reference value, and median(d) represents the median of the last N training data.

[0023] Preferably, setting the temperature gradient threshold value reference value comprises:

[0024] setting an initial value of the temperature gradient threshold value; and obtaining age information and medical history information of the patient;

[0025] adjusting the temperature gradient threshold value according to the age of the patient, comprising: in response to the patient's age being greater than a preset age threshold, increasing the temperature gradient threshold value by 10%, to obtain a first temperature gradient threshold value;

[0026] adjusting the temperature gradient threshold value according to the medical history of the patient, comprising: in response to the patient having diabetes, increasing the first temperature gradient threshold value by 20%, to obtain the temperature gradient threshold value reference value.

[0027] Preferably, it further comprises long-term stage adjustment of the temperature gradient threshold value, comprising:

[0028] obtaining the rehabilitation stage currently undergone by the patient;

[0029] in response to the patient being in the acute stage, setting the temperature gradient threshold value to be not greater than 0.2°C; and in response to the patient being in the recovery stage, increasing the temperature gradient threshold value by 5% every week.

[0030] Preferably, it further comprises long-term self-adaptive adjustment of the electromyographic signal amplitude threshold value, comprising: obtaining the rehabilitation stage currently undergone by the patient; and in response to the patient being in the consolidation stage, setting the electromyographic signal amplitude threshold value to be 80% of the electromyographic signal amplitude of the healthy side of the patient.

[0031] Preferably, it further comprises, before obtaining the original thermal distribution map, compensating the temperature collected by the infrared thermal imager, comprising:

[0032] obtaining the temperature of the environmental heat source;

[0033] obtaining the compensated temperature by using a multispectral data fusion algorithm, and the calculation expression is:

[0034] P1 = P - β × P2;

[0035] wherein P1 and P respectively represent the compensated temperature, P2 represents the temperature of the environmental heat source, and β is a weight coefficient.

[0036] In a second aspect, the present application provides an evaluation and training system based on a medical infrared thermal imager, comprising a memory and a processor, wherein the memory stores computer program instructions, and the computer program instructions, when executed by the processor, implement the evaluation and training method based on the medical infrared thermal imager of the present application.

[0037] The beneficial effects of the present application are that: generally, when the target area of the human body has an abnormality, the temperature gradient and the myoelectric signal amplitude of the target area will be larger, the method of the present application comprehensively considers the temperature gradient and the myoelectric signal amplitude of the target area when detecting whether the target area of the human body has an abnormality, and combines the thermal map and the myoelectric signal to perform double-threshold judgment on the target area, so as to realize the collaborative analysis of biological signals and avoid misjudgment; when it is detected that the target area has an abnormality, the trained neural network model is further used to classify and evaluate the abnormality of the abnormal area, so as to more accurately evaluate the abnormal type and severity; the efficiency of formulating a rehabilitation training scheme is effectively improved by automatically matching the training scheme, and when the training scheme is matched, the current physiological state of the target area is automatically matched, instead of using a fixed training scheme, so as to avoid the adverse consequences caused by overtraining or insufficient training intensity, and help to improve the rehabilitation training effect of the patient.

[0038] Further, by introducing a priority mechanism, the dominant cause can be clearly determined, and the most critical training action and intensity can be preferentially matched, so as to realize focused intervention and improve the pertinence and controllability of the training effect.

[0039] Further, by performing short-term self-adaptive adjustment on the temperature gradient threshold, misjudgment can be further avoided, so as to improve the accuracy of the evaluation results of the abnormal type and severity. BRIEF DESCRIPTION OF DRAWINGS

[0040] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description read in conjunction with the accompanying drawings. In the drawings, several embodiments of the present application are shown by way of example and not limitation, and the same or corresponding reference numbers indicate the same or corresponding parts, in which:

[0041] Figure 1 is a flow chart of an evaluation and training method based on a medical infrared thermal imager according to an embodiment of the present application;

[0042] Figure 2 is a structural schematic diagram of an evaluation and training system based on a medical infrared thermal imager according to an embodiment of the present application. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0044] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0045] Example of an evaluation and training method based on a medical infrared thermal imager:

[0046] like Figure 1 As shown, the evaluation and training method based on the medical infrared thermal imager of the present invention includes:

[0047] S1. Obtaining the original thermal distribution map and surface electromyographic signal of the target area, specifically: using a medical infrared thermal imager to dynamically scan the target area to obtain the original thermal distribution map; synchronously obtaining the surface electromyographic signal of the target area through an attached sensor, and aligning it with the thermal imaging data in time and space;

[0048] The medical infrared thermal imager may be a high-precision medical infrared thermal imager with a resolution of ≤ 0.05°C. The target area may be a joint or muscle group.

[0049] The deviation between the acquisition time of surface electromyography signal and the acquisition time of thermal imaging data must be less than 5ms.

[0050] S2. De-noising the original thermal distribution map using a wavelet transform algorithm to obtain a de-noised thermal distribution map;

[0051] By denoising, the signal-to-noise ratio of medical infrared thermal imagers can be improved to above 30dB.

[0052] S3. Obtaining the temperature gradient of the target area, specifically: obtaining the temperature gradient of the target area based on the de-noised thermal distribution map; the temperature gradient is used to represent the temperature change rate between the target area and the surrounding tissue;

[0053] S4, determining whether the temperature gradient of the target area is greater than a temperature gradient threshold, and whether the amplitude of the electromyographic signal is less than a myoelectric signal amplitude threshold;

[0054] S5. Obtain the abnormal classification result and the corresponding severity rating, specifically: in response to the temperature gradient of the target area being greater than the temperature gradient threshold and the amplitude of the electromyographic signal being less than the electromyographic signal amplitude threshold, the denoised thermal distribution map and the surface electromyographic signal are input into the neural network model, thereby obtaining the abnormal classification result and the corresponding severity rating of the target area.

[0055] In this embodiment, the types of abnormal classification results include inflammation, muscle strain, and nerve damage; the types of severity ratings include level 1, level 2, level 3, level 4, level 5, and level 6.

[0056] S6、According to the abnormal classification result and the corresponding severity rating, the corresponding training type and training intensity are matched.

[0057] In this embodiment, for inflammation of 1-3 levels, the corresponding training type is passive joint movement (without weight), and the corresponding training intensity is 2 groups per day, 5 minutes per group. For inflammation of 4-6 levels, the corresponding training type is low resistance elastic band training, and the corresponding training intensity is resistance of 15-20 pounds and knee angle less than or equal to 45 degrees. For muscle strain of 1-3 levels, the corresponding training type is isometric contraction training, and the training intensity is 5 seconds per contraction and 10 times per group. For nerve damage of 4 levels or more, the corresponding training type is neuromuscular electrical stimulation + mirror training, and the training intensity is electrical stimulation intensity of 20 mA.

[0058] In this embodiment, for inflammation of 4-6 levels, the calculation expression of training resistance f is:

[0059]

[0060] In the formula, f b represents the basic resistance, and N represents the level number.

[0061] Generally, when there is an abnormality in the target region of the human body, the temperature gradient and the electromyographic signal amplitude of the target region will be larger. The method of the present application considers both the temperature gradient and the electromyographic signal amplitude of the target region when detecting whether there is an abnormality in the target region of the human body, and performs double-threshold judgment on the target region by combining the thermogram and the electromyographic signal, thereby realizing collaborative analysis of biological signals and avoiding misjudgment. When it is detected that there is an abnormality in the target region, the trained neural network model is further used to classify and evaluate the abnormality of the abnormal region, thereby more accurately evaluating the type and severity of the abnormality. When matching the training scheme, the current physiological state of the target region is automatically matched, instead of using a fixed training scheme, thereby avoiding the adverse consequences caused by overtraining or insufficient training intensity, and helping to improve the rehabilitation training effect of the patient.

[0062] In one embodiment, if the abnormal type output by the neural network model is multiple, the training action and training intensity corresponding to the abnormal type with higher priority are selected according to a preset abnormal type priority order; the preset abnormal type priority order is inflammation > nerve damage > muscle strain.

[0063] For example, if inflammation (level 3) and muscle damage (level 2) are detected at the same time, the inflammation training scheme is preferentially matched.

[0064] By setting inflammation as the highest priority, the worsening of inflammation can be avoided. By setting the abnormal type with higher priority, it is helpful to preferentially train the abnormal type with higher priority, thereby improving the training effect.

[0065] In one embodiment, the training action and training intensity are also regionally adjusted, and the adjustment rules include: if the abnormal region is below the knee patella, deep squatting is prohibited; and if inflammation occurs in the rotator cuff muscle group, the abduction angle is limited to less than or equal to 60 degrees.

[0066] In one embodiment, the training method of the neural network model includes:

[0067] S101, acquire a data set and label and preprocess it to form a training set, the data set including multiple target region infrared thermal imaging images and corresponding SEMG data, the multiple target region infrared thermal imaging images including healthy target region infrared thermal imaging images, target region infrared thermal imaging images with inflammation, target region infrared thermal imaging images with muscle strain, and target region infrared thermal imaging images with nerve damage.

[0068] In this embodiment, there are 1000 target region infrared thermal imaging images. The labels for labeling include normal, inflammation, muscle strain, and nerve damage.

[0069] In this embodiment, the preprocessing includes:

[0070] 1) image standardization, including: linearly normalizing the temperature of the original thermal image in the range of [0, 1], the formula being as follows:

[0071]

[0072] In the formula, T norm represents the normalized temperature, T min and T max respectively represent the highest temperature and the lowest temperature, and T represents the temperature before normalization.

[0073] The resolution of the original thermal image is adjusted to 224x224.

[0074] By adjusting the resolution to 224x224, the input size of the ResNet network can be adapted.

[0075] 2) data augmentation, including: random horizontal inversion and brightness adjustment of the original thermal image;

[0076] The horizontal inversion can simulate the left-right symmetrical parts, and the brightness adjustment is in the form of increasing the brightness by 10% or decreasing the brightness by 10%, which can simulate environmental noise through brightness adjustment.

[0077] S102, input layer adaptation and output layer modification are performed on the ResNet-18 model to obtain a neural network model, the input layer adaptation includes: modifying the 3 channels of the input layer to a single channel; the output layer modification includes: replacing the original 1000-class classification layer with a custom output head, the custom output head including three fully connected layers and a softmax classifier, wherein the three fully connected layers correspond to three types of abnormalities respectively, i.e., inflammation, muscle strain and nerve injury.

[0078] S103, setting a loss function, setting an optimizer, hyperparameters and regularization.

[0079] In this embodiment, the loss function is set as:

[0080] L=α.L CE +(1-α).L MSE

[0081] In the formula, L represents the total loss function, α represents the hyperparameter, which controls the weight distribution between the classification loss and the regression loss. L CE represents the classification loss, L MSE represents the regression loss.

[0082] In this embodiment, the optimizer is AdamW optimizer, the initial learning rate is set to e -4 , the batch size is set to 32, and the training round is set to 50.

[0083] In this embodiment, the regularization includes: adding Dropout before the fully connected layer, and label balancing processing, the label smoothing coefficient is 0 or 1, so as to prevent overfitting.

[0084] S104, iteratively training the neural network model, including: forward propagation, loss calculation, backward propagation and parameter updating.

[0085] In one embodiment, it also includes short-term adaptive adjustment of the temperature gradient threshold, including:

[0086] S201, setting a temperature gradient threshold reference value;

[0087] S202, obtaining data of the last N times of training of the patient, and calculating the updated temperature gradient threshold, the data of a certain training refers to the temperature of the target area after the training; the updated temperature gradient threshold calculation expression is:

[0088] T new =∝×T0+(1-∝)×median(d);

[0089] In the formula, T newrepresents an updated temperature gradient threshold value, a is a forgetting factor, T0 represents a temperature gradient threshold value reference value, and median(d) represents the median of the last N training data.

[0090] Since the temperature gradient of the target region and the amplitude of the electromyographic signal are affected by age, medical history, and rehabilitation stage, even if the target region is normal, the temperature gradient may be too large or the amplitude of the electromyographic signal may be too large. By short-term self-adaptive adjustment of the temperature gradient threshold value, the occurrence of misjudgment can be further avoided, thereby improving the accuracy of the evaluation result of the abnormal type and severity.

[0091] In this embodiment, the value of a is in the range of 0.7-0.9.

[0092] In one embodiment, setting the temperature gradient threshold value reference value comprises:

[0093] S301, setting a temperature gradient threshold value initial value; and obtaining age information and medical history information of the patient;

[0094] In this embodiment, the temperature gradient threshold value T is set to 0.3℃.

[0095] S302, adjusting the temperature gradient threshold value according to the age of the patient, comprising: in response to the patient's age being greater than a preset age threshold, increasing the temperature gradient threshold value by 10%, to obtain a first temperature gradient threshold value;

[0096] After increasing by 10%, the temperature gradient threshold value is increased from 0.3℃ to 0.33℃.

[0097] S303, adjusting the temperature gradient threshold value according to the medical history of the patient, comprising: in response to the patient having diabetes, increasing the first temperature gradient threshold value by 20%, to obtain a temperature gradient threshold value reference value.

[0098] The method of this embodiment comprehensively considers the influence of age and diabetes on the temperature gradient and electromyographic signal of the target region when setting the temperature gradient reference value, so that the temperature gradient reference value is more scientific and reasonable.

[0099] In one embodiment, it further comprises long-term stage adjustment of the temperature gradient threshold value, comprising:

[0100] S401, obtaining the rehabilitation stage currently undergone by the patient;

[0101] S402, in response to the patient being in the acute stage, setting the temperature gradient threshold value to be not greater than 0.2℃; and in response to the patient being in the recovery stage, increasing the temperature gradient threshold value by 5% every week.

[0102] In the acute phase (such as the early stage of inflammation, the early stage after operation), the temperature gradient threshold value is set to a lower value (≤0.2℃), which can more sensitively capture the slight abnormal heat change and prevent the lesion area from deteriorating due to untimely discovery; in the recovery stage, the threshold value is gradually increased (increased by 5% per week), to avoid false positives caused by excessive sensitivity, and to improve the robustness and fault tolerance of the system, while more reasonably distinguishing between physiological recovery-induced heat variation and real abnormalities. In addition, patients in different rehabilitation stages have large differences in basal metabolism, microcirculation, and heat diffusion mechanisms, and a fixed threshold value is easy to cause evaluation errors; the present scheme has good individual adaptability and time sequence adaptability, and can significantly reduce the probability of misjudgment or omission.

[0103] In one embodiment, the method further comprises long-term adaptive adjustment of the electromyographic signal amplitude threshold value, comprising: obtaining the rehabilitation stage in which the patient is currently located; and in response to the patient being in the consolidation stage, setting the electromyographic signal amplitude threshold value to 80% of the electromyographic signal amplitude of the patient's healthy side.

[0104] In one embodiment, the temperature gradient of the target area comprises:

[0105] S501, obtaining a thermogram matrix of a target area;

[0106] S502, calculating the gradients in the x direction and the y direction of the thermogram matrix, and the calculation expression is:

[0107]

[0108] In the above two expressions, G x and G y respectively represent the gradients in the x direction and the y direction, and T represents the thermogram matrix of the target area.

[0109] S503, calculating the total gradient amplitude G, and the calculation expression is:

[0110]

[0111] S504, normalizing the total gradient amplitude G.

[0112] In one embodiment, the method further comprises, before obtaining the original thermogram, compensating the temperature collected by the infrared thermal imager, comprising:

[0113] S601, obtaining the temperature of the environmental heat source;

[0114] S602, obtaining the compensated temperature by using a multispectral data fusion algorithm, and the calculation expression is:

[0115] P1=P-β×P2;

[0116] Where P1 and P represent the compensated temperatures, P2 represents the temperature of the ambient heat source, and β is the weight coefficient.

[0117] The stability of thermal imaging data can be ensured by compensating the temperature collected by the infrared thermal imager.

[0118] Example of an evaluation and training system based on a medical infrared thermal imager:

[0119] The present invention also provides an evaluation and training system based on a medical infrared thermal imager. Figure 2 As shown, the evaluation and training system based on a medical infrared thermal imager includes a processor and a memory, and the memory stores computer program instructions. When the computer program instructions are executed by the processor, an evaluation and training method based on a medical infrared thermal imager according to the first aspect of the present invention is implemented.

[0120] The medical infrared thermal imager-based assessment and training system also includes other components familiar to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

[0121] While several embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous modifications, variations, and alternatives will occur to those skilled in the art without departing from the concept and spirit of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

Claims

1. An evaluation and training method based on a medical infrared thermal imager, characterized in that: include: Use a medical infrared thermal imager to dynamically scan the target area to obtain the original thermal distribution map; The surface electromyographic signals of the target area are acquired synchronously through attached sensors and aligned with the thermal imaging data in time and space. The original thermal distribution map is denoised using a wavelet transform algorithm to obtain a denoised thermal distribution map; The temperature gradient of the target area is obtained based on the denoised thermal distribution map; the temperature gradient is used to characterize the temperature change rate between the target area and the surrounding tissue; Determine whether the temperature gradient of the target area is greater than the temperature gradient threshold and whether the amplitude of the electromyographic signal is less than the electromyographic signal amplitude threshold; In response to the temperature gradient of the target area being greater than the temperature gradient threshold and the amplitude of the electromyographic signal being less than the electromyographic signal amplitude threshold, the denoised thermal distribution map and the surface electromyographic signal are input into the neural network model to obtain an abnormality classification result and a corresponding severity rating of the target area; Match the corresponding training type and training intensity based on the abnormal classification results and the corresponding severity rating.

2. The evaluation and training method based on a medical infrared thermal imager according to claim 1, characterized in that: If the neural network model outputs multiple abnormality types, the training action and training intensity corresponding to the abnormality type with higher priority are selected according to the preset abnormality type priority order; The preset priority order of abnormality types is inflammation > nerve damage > muscle strain.

3. The evaluation and training method based on a medical infrared thermal imager according to claim 1, wherein: It also includes regional adaptive adjustments to training movements and training intensity. The adjustment rules include: if the abnormal area is below the patella, squats are prohibited; if inflammation occurs in the rotator cuff muscles, the abduction angle is limited to less than or equal to 60 degrees.

4. The evaluation and training method based on a medical infrared thermal imager according to claim 1, wherein: The training methods for neural network models include: Acquiring a data set, annotating and preprocessing the data set to form a training set, wherein the data set includes a plurality of infrared thermal images of a target area and corresponding SEMG data, wherein the plurality of infrared thermal images of the target area include infrared thermal images of a healthy target area, infrared thermal images of a target area with inflammation, infrared thermal images of a target area with muscle strain, and infrared thermal images of a target area with nerve damage; The ResNet-18 model was adapted for the input layer and modified for the output layer to obtain a neural network model. The input layer adaptation included changing the three channels of the input layer to a single channel. The output layer modification included replacing the original 1000-class classification layer with a custom output head. The custom output head consisted of three fully connected layers and a softmax classifier. The three fully connected layers corresponded to three abnormality types: inflammation, muscle strain, and nerve damage. Set the loss function, optimizer, hyperparameters, and regularization; The neural network model is trained iteratively, including forward propagation, loss calculation, backpropagation, and parameter update.

5. The evaluation and training method based on a medical infrared thermal imager according to claim 1, wherein: Also included is short-term adaptive adjustment of temperature gradient thresholds, including: Set the temperature gradient threshold reference value; Obtain the patient's most recent N training data and calculate the updated temperature gradient threshold. The data of a certain training session refers to the temperature of the target area after that training session. The calculation expression for the updated temperature gradient threshold is: T new =∝×T0+(1-∝)×median(d); Where, T new represents the updated temperature gradient threshold, ∝ is the forgetting factor, T0 represents the temperature gradient threshold reference value, and median(d) represents the median of the data of the last N training times.

6. The evaluation and training method based on a medical infrared thermal imager according to claim 5, characterized in that: Setting the temperature gradient threshold reference value includes: Set the initial value of the temperature gradient threshold; and obtain the patient's age information and medical history information; Adjusting the temperature gradient threshold according to the patient's age includes: in response to the patient's age being greater than a preset age threshold, increasing the temperature gradient threshold by 10% to obtain a first temperature gradient threshold; The temperature gradient threshold is adjusted according to the patient's medical history, including: in response to the patient having diabetes, the first temperature gradient threshold is increased by 20%, thereby obtaining a temperature gradient threshold reference value.

7. The evaluation and training method based on a medical infrared thermal imager according to claim 1, wherein: Also included are long-term phase adjustments to temperature gradient thresholds, including: Obtain the patient's current stage of recovery; In response to the patient being in the acute phase, the temperature gradient threshold is set to no more than 0.2° C.; in response to the patient being in the recovery phase, the temperature gradient threshold is increased by 5% every other week.

8. The evaluation and training method based on a medical infrared thermal imager according to claim 1, wherein: The method also includes long-term adaptive adjustment of the electromyographic signal amplitude threshold, including: obtaining the patient's current rehabilitation stage; and setting the electromyographic signal amplitude threshold to 80% of the electromyographic signal amplitude of the patient's healthy side in response to the patient being in the consolidation period.

9. The evaluation and training method based on a medical infrared thermal imager according to any one of claims 1 to 8, characterized in that: Also includes: Before obtaining the original thermal distribution map, the temperature collected by the infrared thermal imager is compensated, including: Get the temperature of the ambient heat source; The multispectral data fusion algorithm is used to obtain the compensated temperature. The calculation expression is: P1=P-β×P2; Where P1 and P represent the compensated temperatures, P2 represents the temperature of the ambient heat source, and β is the weight coefficient.

10. An assessment and training system based on a medical infrared thermal imager, comprising a memory and a processor, wherein the memory stores computer program instructions, characterized in that: When the computer program instructions are executed by the processor, the evaluation and training method based on a medical infrared thermal imager according to any one of claims 1 to 9 is implemented.

Images