An intelligent ICF-RS function evaluation grading system and method

Abstract

The present invention discloses an intelligent ICF-RS functional assessment and grading system and method, which specifically includes an input module, different dimensional assessment modules, and an output module; the different dimensional assessment modules include: a physical function dimension assessment module, an activity function dimension assessment module, a participation function dimension assessment module, and an overall functional status assessment module. The outputs of the input modules are respectively input into the different dimensional assessment modules, and the outputs of the different dimensional assessment modules are then input into the output module and output the final functional status grading results. Compared with the traditional clinical technology that applies the ICF evaluation standard, this method uses a large amount of ICF-RS quantitative standard functional status grading data to train and optimize the four-dimensional neural network model, and can evaluate the patient's functional status in 30 categories separately, and at the same time output the four-dimensional functional grade assessment results, so as to obtain more objective diagnostic results, reduce the misdiagnosis rate, and reduce the patient's medical costs.

Description

Technical Field

[0001] The present invention relates to the field of computer-assisted medical technology, and more particularly to an intelligent ICF-RS functional assessment grading system and method. Background Art

[0002] The International Classification of Functioning, Disability, and Health (ICF-RS) consists of 30 items. The ICF-RS can be used to assess functional status, such as health and impairment. However, when using the ICF-RS to assess functional status, the criteria for categorizing each item must follow the category limits specified in the World Health Organization's ICF Operational Manual. Because the item limits (limit 0 corresponds to 1%-4%, limit 1 corresponds to 5%-24%, limit 2 corresponds to 25%-49%, limit 3 corresponds to 50%-95%, and limit 4 corresponds to 96%-100%) are unequally spaced, it is impossible to aggregate the results of all 30 items. This means that when explaining functional changes to patients, one must interpret each category individually, failing to provide an assessment of the three dimensions of physical function, activity, and participation, or overall functional status. Consequently, functional outcomes cannot be accurately assessed in clinical practice, severely hindering the clinical application of the ICF-RS.

[0003] At present, the only way to solve this problem is to use other assessment criteria instead of the limit value as the evaluation result of each item. For example: (1) Use the Numerical Rating Scale (NRS) as the calibration standard for ICF items; (2) Match the items of existing clinical tools with the ICF-RS items and use the evaluation results of existing clinical tools as the evaluation standard for ICF-RS items. For example, "eating" in the Barthel Index (BI) assessment scale corresponds to the ICF-RS item d550 eating, and the BI scoring standard is applied during the evaluation. However, the first method violates the original intention of ICF and is different from the internationally accepted ICF evaluation standard; the second method also only has some ICF items that can be matched with existing clinical evaluation tools. Moreover, if different items are matched with different clinical evaluation tools, the evaluation results may still not be summarized due to the inconsistent evaluation standards of different clinical evaluation tools.

[0004] The prior art discloses a rehabilitation information management system and method based on the International Classification of Functioning, Disability, and Health. This method proposes an ICF-RS information management system for continuous monitoring and electronic recording of rehabilitation patients' functional levels, enabling horizontal and vertical comparison, analysis, and prediction of functional information across different medical institutions, diseases, and time periods. However, this approach suffers from a flaw: it fails to iteratively train the neural network model using ICF information and expert assessment results, preventing the generation of summary results for the three dimensions and overall functional level assessment.

[0005] To this end, in combination with the above requirements and the shortcomings of the existing technology, this application proposes an intelligent ICF-RS function evaluation and grading system and method. Summary of the Invention

[0006] The present invention provides an intelligent ICF-RS functional assessment grading system and method, which can evaluate the functional status of patients separately with reference to the ICF standard, and can simultaneously summarize the physical function dimension level assessment results, activity function dimension level assessment results, participation function dimension level assessment results, and overall functional status level assessment results.

[0007] The primary purpose of the present invention is to solve the above technical problems, and the technical solutions of the present invention are as follows:

[0008] The first aspect of the present invention provides an intelligent ICF-RS functional assessment and grading system, comprising an input module, different dimensional assessment modules, and an output module; the outputs of the input modules are respectively input into the different dimensional assessment modules, and the outputs of the different dimensional assessment modules are then input into the output module and output a final functional status grading result; the functional status grading results output by the output module include: normal, mild functional impairment, moderate functional impairment, and severe functional impairment.

[0009] Furthermore, the different dimension assessment modules include four different dimensions, and the four different dimensions are divided according to ICF-RS items. The different dimension assessment modules include: a physical function dimension assessment module, an activity function dimension assessment module, a participation function dimension assessment module, and an overall functional status assessment module. The physical function dimension assessment module is used to assess ICF-RS items related to physical function, the activity function dimension assessment module is used to assess ICF-RS items related to activity function, the participation function dimension assessment module is used to assess ICF-RS items related to participation function, and the overall functional status assessment module is used to assess all ICF-RS items.

[0010] A second aspect of the present invention provides an intelligent ICF-RS function evaluation and grading method, which is used in the intelligent ICF-RS function evaluation and grading system, comprising:

[0011] S1. Collect the original data of ICF-RS quantitative standard functional status classification, pre-process the original data and divide the data into training set, validation set and test set.

[0012] S2. Construct neural network models corresponding to four different dimensions, use the training set to train the four neural network models respectively, use the loss function to calculate the loss value, use the back propagation technology to update the network parameters of the model, complete a round of training, and obtain the preliminary optimal model.

[0013] S3. Input the validation set into the preliminary optimal model to obtain a preliminary prediction result, determine whether the preliminary prediction result is consistent with the label value, calculate the accuracy of the current model on the validation set, and determine whether the accuracy is greater than the set value and is the highest. If so, save the model as the current optimal model, otherwise repeat the training process of step S2;

[0014] S4. Use the test set to test the current optimal model, obtain the function level evaluation performance result of the current optimal model on the test set, and judge whether the result is consistent with the true label value. If so, determine the current optimal model, otherwise repeat step S3.

[0015] S5. Collect the patient's ICF-RS quantitative standard functional status data, input it into the saved current optimal model after preprocessing, and output the patient's functional level assessment in four dimensions.

[0016] The label value is a "label value" determined manually by multiple experts and has several deviations.

[0017] Among them, the backpropagation algorithm consists of two cyclic iterations: error propagation and parameter update. During the backpropagation process of machine learning, the various parameters of the loss function are guided to update to the appropriate size in the correct direction, so that the updated parameters make the loss function value continuously approach the global minimum.

[0018] Furthermore, the ICF-RS quantitative standard functional status grading original data in step S1 includes the evaluation results of multiple experts, specifically: 30 ICF category evaluation results, physical function status evaluation results, activity function status evaluation results, participation function status evaluation results, and overall function status evaluation results.

[0019] Furthermore, the process of preprocessing the raw data in step S1 includes:

[0020] S11. Eliminate data with missing values ​​or unreasonable data.

[0021] S12. Map the four labels of normal, mild dysfunction, moderate dysfunction, and severe dysfunction in the four dimensions to four preset numbers.

[0022] S13. Using the majority selection rule, the evaluation results of multiple experts on a sample data are used as the label of the sample.

[0023] The purpose of removing missing values ​​or unreasonable data is to reduce the impact of human errors on neural network training.

[0024] Among them, when collecting ICF-RS data, the evaluation results of multiple experts on the same patient are collected. The results of multiple experts may be inconsistent, so the mode is used as the final result.

[0025] Furthermore, the neural network models corresponding to the four different dimensions include: physical function dimension model, activity function dimension model, participation function dimension model, and overall function dimension model, among which the data set used by the physical function dimension model includes 9 category data on physical function dimensions in ICF, as well as physical function status assessment results; the data set used by the activity function dimension model includes 14 category data on daily activity dimensions in ICF, as well as activity function status assessment results; the data set used by the participation function dimension model includes 7 category data on social participation dimensions in ICF, as well as participation function status assessment results; the data set used by the overall function dimension model includes 30 category data in ICF-RS and overall function status assessment results; finally, the data sets are divided into training set, validation set, and test set according to the set ratio.

[0026] Furthermore, the process of constructing neural network models corresponding to four different dimensions is specifically as follows: the grid search method is used to determine the number of neuron nodes and network layers of the four neural network models, wherein the overall function dimension model inputs 30 category data in ICF-RS, and the output is the overall function level assessment result; the body function dimension model inputs 9 category data of body function dimensions in ICF, and the output is the body function level assessment result; the activity function dimension model inputs 14 category data of daily activity dimensions in ICF, and the output is the activity function level assessment result; the participation function dimension model inputs 7 category data of social participation dimensions in ICF, and the output is the participation function level assessment result.

[0027] Furthermore, the loss functions of the neural network models corresponding to the four different dimensions are all calculated using cross-entropy loss, and the stochastic gradient optimizer SGD is used to update and calculate the network parameters that affect the model training and output, so that the network parameters approach the optimal value, thereby minimizing the loss function.

[0028] Among them, the back propagation technology calculates the preliminary neural network parameters according to the loss of each round, and the stochastic gradient optimizer SGD is used to fine-tune these parameters to achieve the purpose of quickly obtaining the optimal network performance.

[0029] Among them, after the network model outputs the result, the cross entropy loss is calculated with the true label, and the network parameters are calculated and updated through the optimizer. The SGD stochastic gradient descent method calculates the gradient direction of the network parameters so that the network parameters change in the opposite direction of the gradient during back propagation, thereby reducing the loss function; the more accurate the network prediction, the smaller the model loss.

[0030] The better the network parameters are designed, the more accurate the network model's output will be, and the smaller the loss (the difference between the model's predicted value and the label value) will be. Minimizing the loss function involves iteratively updating the network parameters to gradually optimize the network and reduce the loss.

[0031] Furthermore, the training process described in step S2 is as follows: the training set of the corresponding dimensional model is used as the input data of the dimensional model, the output result is expressed as the predicted probability of the functional level result, the output result is compared with the label value to calculate the loss value, and the stochastic gradient optimizer SGD is used to update and calculate the network parameters that affect the model training and output, so that the network parameters approach the optimal value, thereby minimizing the loss function and obtaining a preliminary optimal model.

[0032] Furthermore, the verification process described in step S3 is as follows: the verification set of the corresponding dimensional model is used as the input data of the dimensional model, the accuracy of the output result compared with the label value is calculated, and when the accuracy is higher than the set value, the model is saved as the current optimal model.

[0033] Among them, the iterative process of the neural network is realized by continuously repeating steps S2 and S3, continuously minimizing the loss value and improving the functional level result evaluation performance of the neural network model.

[0034] Furthermore, the collected ICF-RS dataset can continue to increase. After assisting medical staff to complete the four-dimensional level assessment of patients, the four-dimensional functional level assessment results obtained can be used as a new dataset to conduct a new round of training for the saved current optimal model.

[0035] Among them, the more ICF-RS data sets collected, the more beneficial it will be for the training of the neural network model.

[0036] Among them, the current optimal model can assist rehabilitation doctors to complete the physical function status level assessment results, activity function status level assessment results, participation function status level assessment results, and overall function status level assessment results faster and better, thereby efficiently determining the patient's diagnosis and treatment plan, reducing the misdiagnosis rate, and improving the quality of medical care enjoyed by patients.

[0037] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0038] The present invention provides an intelligent ICF-RS functional assessment and grading system and method. By using the ICF-RS quantitative standard functional status grading data, neural network models corresponding to four different dimensions are trained and optimized. The system can evaluate the patient's functional status in 30 categories separately and output the functional grade assessment results of the four dimensions at the same time, thereby obtaining more objective diagnostic results, reducing the misdiagnosis rate, and lowering the patient's medical costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a structural diagram of an intelligent ICF-RS function evaluation and grading system of the present invention.

[0040] Figure 2 This is a flow chart of an intelligent ICF-RS function evaluation and grading method of the present invention.

[0041] Figure 3 Schematic diagram of the body function dimensional model in the present invention.

[0042] Figure 4 This is a schematic diagram of the activity function dimensional model in the present invention.

[0043] Figure 5 This is a schematic diagram of the functional dimension model involved in the present invention.

[0044] Figure 6 It is a schematic diagram of the overall functional dimension model in the present invention.

[0045] Figure 7 This is a flowchart of training a single neural network model in the present invention.

[0046] Figure 8 This is a diagram of the overall steps for training a neural network in the present invention.

[0047] Figure 9 This is a functional evaluation classification performance comparison table of an embodiment of the present invention.

[0048] Figure 10 This is a functional evaluation classification performance comparison table of another embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0050] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0051] Example 1

[0052] like Figure 1As shown, the present invention provides an intelligent ICF-RS functional assessment and grading system, including an input module, different dimensional assessment modules, and an output module; the outputs of the input modules are respectively input into the different dimensional assessment modules, and the outputs of the different dimensional assessment modules are then input into the output module and output a final functional status grading result; the functional status grading results output by the output module include: normal, mild functional impairment, moderate functional impairment, and severe functional impairment.

[0053] Furthermore, the different dimension assessment modules include four different dimensions, and the four different dimensions are divided according to ICF-RS items. The different dimension assessment modules include: a physical function dimension assessment module, an activity function dimension assessment module, a participation function dimension assessment module, and an overall functional status assessment module. The physical function dimension assessment module is used to assess ICF-RS items related to physical function, the activity function dimension assessment module is used to assess ICF-RS items related to activity function, the participation function dimension assessment module is used to assess ICF-RS items related to participation function, and the overall functional status assessment module is used to assess all ICF-RS items.

[0054] Among them, by collecting the limit value results of 30 items in the ICF-RS of clinically sampled patients and the functional grade results of four dimensions, the patient's functional status can be evaluated separately in 30 categories, and the three dimensions and the overall functional grade results can be summarized at the same time, which solves the problem that the ICF-RS limit values ​​cannot be summarized, and also retains the application of category limit values.

[0055] Example 2

[0056] Based on the above embodiment 1, combined Figure 2-Figure 8 This embodiment describes in detail an intelligent ICF-RS function evaluation and grading method for the intelligent ICF-RS function evaluation and grading system described in Example 1, such as Figure 2 As shown, including:

[0057] S1. Collect the original data of ICF-RS quantitative standard functional status classification, pre-process the original data and divide the data into training set, validation set and test set.

[0058] S2. Construct neural network models corresponding to four different dimensions, use the training set to train the four neural network models respectively, use the loss function to calculate the loss value, use the back propagation technology to update the network parameters of the model, complete a round of training, and obtain the preliminary optimal model.

[0059] S3. Input the validation set into the preliminary optimal model to obtain a preliminary prediction result, determine whether the preliminary prediction result is consistent with the label value, calculate the accuracy of the current model on the validation set, and determine whether the accuracy is greater than the set value and is the highest. If so, save the model as the current optimal model, otherwise repeat the training process of step S2;

[0060] S4. Use the test set to test the current optimal model, obtain the function level evaluation performance result of the current optimal model on the test set, and judge whether the result is consistent with the true label value. If so, determine the current optimal model, otherwise repeat step S3.

[0061] S5. Collect the patient's ICF-RS quantitative standard functional status data, input it into the saved current optimal model after preprocessing, and output the patient's functional level assessment in four dimensions.

[0062] The label value is a "label value" determined manually by multiple experts and has several deviations.

[0063] Among them, the backpropagation algorithm consists of two cyclic iterations: error propagation and parameter update. During the backpropagation process of machine learning, the various parameters of the loss function are guided to update to the appropriate size in the correct direction, so that the updated parameters make the loss function value continuously approach the global minimum.

[0064] In a specific embodiment, the ratio of the training set, the validation set, and the test set is 8:2:2.

[0065] Furthermore, the ICF-RS quantitative standard functional status grading original data in step S1 includes the evaluation results of multiple experts, specifically: 30 ICF category evaluation results, physical function status evaluation results, activity function status evaluation results, participation function status evaluation results, and overall function status evaluation results.

[0066] The evaluation results of the 30 ICF categories are shown in the following table:

[0067]

[0068]

[0069] Furthermore, the process of preprocessing the raw data in step S1 includes:

[0070] S11. Eliminate data with missing values ​​or unreasonable data.

[0071] S12. Map the four labels of normal, mild dysfunction, moderate dysfunction, and severe dysfunction in the four dimensions to four preset numbers.

[0072] S13. Using the majority selection rule, the evaluation results of multiple experts on a sample data are used as the label of the sample.

[0073] Among them, when collecting ICF-RS data, the evaluation results of multiple experts on the same patient are collected. The results of multiple experts may be inconsistent, so the mode is used as the final result.

[0074] The purpose of eliminating missing values ​​or unreasonable data is to reduce the impact of human errors on neural network training. In a specific embodiment, when multiple doctors evaluate a patient, one doctor evaluates item 9, while other doctors evaluate it as 0 to 4, the data of item 9 needs to be eliminated.

[0075] Furthermore, the neural network models corresponding to the four different dimensions include: physical function dimension model, activity function dimension model, participation function dimension model, and overall function dimension model, among which the data set used by the physical function dimension model includes 9 category data on physical function dimensions in ICF, as well as physical function status assessment results; the data set used by the activity function dimension model includes 14 category data on daily activity dimensions in ICF, as well as activity function status assessment results; the data set used by the participation function dimension model includes 7 category data on social participation dimensions in ICF, as well as participation function status assessment results; the data set used by the overall function dimension model includes 30 category data in ICF-RS and overall function status assessment results; finally, the data sets are divided into training set, validation set, and test set according to the set ratio.

[0076] Furthermore, the process of constructing the neural network models corresponding to the four different dimensions is as follows: the grid search method is used to determine the number of neuron nodes and network layers of the four neural network models, wherein the overall functional dimension model input is the 30 category data in ICF-RS, and the output is the overall functional level evaluation result. The specific structure is as follows Figure 6 As shown in the figure, the input of the physical function dimension model is the category data of the 9 physical function dimensions in ICF, and the output is the physical function level assessment result. The specific structure is as follows: Figure 3 As shown in the figure, the input of the activity function dimension model is the category data of the 14 daily activity dimensions in ICF, and the output is the activity function level evaluation result. The specific structure is as follows Figure 4 As shown in the figure, the input of the participation function dimension model is the category data of the 7 social participation dimensions in ICF, and the output is the participation function level evaluation result. The specific structure is as follows: Figure 5 shown.

[0077] Furthermore, the loss functions of the neural network models corresponding to the four different dimensions are all calculated using cross-entropy loss, and the stochastic gradient optimizer SGD is used to update and calculate the network parameters that affect the model training and output, so that the network parameters approach the optimal value, thereby minimizing the loss function.

[0078] Among them, the back propagation technology calculates the preliminary neural network parameters according to the loss of each round, and the stochastic gradient optimizer SGD is used to fine-tune these parameters to achieve the purpose of quickly obtaining the optimal network performance.

[0079] Among them, after the network model outputs the result, the cross entropy loss is calculated with the true label, and the network parameters are calculated and updated through the optimizer. The SGD stochastic gradient descent method calculates the gradient direction of the network parameters so that the network parameters change in the opposite direction of the gradient during back propagation, thereby reducing the loss function; the more accurate the network prediction, the smaller the model loss.

[0080] The better the network parameters are designed, the more accurate the network model's output will be, and the smaller the loss (the difference between the model's predicted value and the label value) will be. Minimizing the loss function involves iteratively updating the network parameters to gradually optimize the network and reduce the loss.

[0081] The loss function is specifically:

[0082]

[0083]

[0084] Among them, N represents the number of samples; K represents the number of categories; y ic ∈{0,1} can be regarded as a one-hot encoding. If the i-th sample belongs to category c, the y at the corresponding position ic Take 1, otherwise take 0; It represents the probability that the i-th sample belongs to category c predicted by the model.

[0085] Furthermore, the training process in step S2 is as follows: Figure 7 As shown in the figure, the specific details are as follows: the training set of the corresponding dimensional model is used as the input data of the dimensional model, the output result is expressed as the predicted probability of the functional level result, the output result is compared with the label value to calculate the loss value, and the stochastic gradient optimizer SGD is used to update and calculate the network parameters that affect the model training and output, so that the network parameters are close to the optimal value, thereby minimizing the loss function and obtaining the preliminary optimal model.

[0086] in, Figure 7Here, best represents the best accuracy of the model in history, i represents the current i-th round of training, acc represents the accuracy of the model's current prediction, and EP represents the total number of epochs of training.

[0087] In a specific embodiment, the training and validation phases are as follows Figure 7 As shown in Figure 2, the steps to optimize the network during one round of training are as follows:

[0088] In step (a), in one round of training, a batch of data is taken for training.

[0089] Step (b): Initialize the parameters, set the current round number i = 1, and the historical best accuracy best = 0.0.

[0090] In step (c), this batch of data is used as the input of the model and the prediction result of the model is output, which is a probability value vector of four dimensions.

[0091] In step (d), the output obtained by the network is compared with the true label and the loss is calculated.

[0092] In step (e), the optimizer SGD is used to update and calculate the network parameters that affect model training and model output, so that they approach or reach the optimal value, thereby minimizing the loss function and finding the optimal model.

[0093] Step (f) After a round of training, the model is validated; all the data in the validation set are put into the model to obtain the prediction results, and the consistency between the prediction results and the true labels is calculated. The accuracy of the current model on the validation set is calculated, and if the current accuracy is the highest, the current model is saved.

[0094] Step (g) determines whether the current round number i is equal to the set total round number EP. If not, repeat steps (a) to (f). If yes, end the training.

[0095] Furthermore, the verification process described in step S3 is as follows: the verification set of the corresponding dimensional model is used as the input data of the dimensional model, the accuracy of the output result compared with the label value is calculated, and when the accuracy is higher than the set value, the model is saved as the current optimal model.

[0096] Among them, the iterative process of the neural network is realized by continuously repeating steps S2 and S3, continuously minimizing the loss value and improving the functional level result evaluation performance of the neural network model.

[0097] In a specific embodiment, Figure 8 As shown, the implementation steps of this method include collecting data, screening unqualified data, data preprocessing, model building, training and validating the model, and testing the accuracy on the optimal model.

[0098] Furthermore, the collected ICF-RS dataset can continue to increase. After assisting medical staff to complete the four-dimensional level assessment of patients, the four-dimensional functional level assessment results obtained can be used as a new dataset to conduct a new round of training for the saved current optimal model.

[0099] Among them, the more ICF-RS data sets collected, the more beneficial it will be for the training of the neural network model.

[0100] Among them, the current optimal model can assist rehabilitation doctors to complete the physical function status level assessment results, activity function status level assessment results, participation function status level assessment results, and overall function status level assessment results faster and better, thereby efficiently determining the patient's diagnosis and treatment plan, reducing the misdiagnosis rate, and improving the quality of medical care enjoyed by patients.

[0101] Example 3

[0102] Based on the above-mentioned embodiment 1 and embodiment 2, this embodiment combines Figure 9-10 The experimental comparison data compares the classification performance of other classifiers in the field, such as Figure 9 As shown in the figure, a 10-fold cross-validation score was used to compare various models during the training phase. The conclusion is as follows: Compared with commonly used classifiers in the field of machine learning, such as GaussianNB, MNB, and CNB, the neural network model used in the present invention performs better in terms of overall function prediction and function prediction in various dimensions, and has a certain degree of stability. In a specific embodiment, as Figure 10 As shown in the figure, another 100 test data were collected and a comparative experiment was conducted. The experimental results are shown in the figure. Figure 10 As shown, the conclusion is: based on the model prediction results of the overall function and the three functional dimensions, the method provided by the present invention has a better performance; other machine learning algorithms cannot be applied well, and the QDA algorithm fails in the participation dimension model.

[0103] The icons in the accompanying drawings that describe the structural positional relationships are for illustrative purposes only and should not be construed as limitations on this patent.

[0104] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. An intelligent ICF-RS function evaluation and grading method, characterized in that: The method is used for an intelligent ICF-RS functional assessment and grading system, which includes an input module, different dimension assessment modules, and an output module; the outputs of the input module are respectively input into the different dimension assessment modules, the different dimension assessment modules include four different dimensions, and the four different dimensions are divided according to ICF-RS items. The different dimension assessment modules include: a body function dimension assessment module, an activity function dimension assessment module, a participation function dimension assessment module, and an overall functional status assessment module, the body function dimension assessment module is used to assess ICF-RS items related to body function, the activity function dimension assessment module is used to assess ICF-RS items related to activity function, the participation function dimension assessment module is used to assess ICF-RS items related to participation function, and the overall functional status assessment module is used to assess all ICF-RS items; the outputs of the different dimension assessment modules are then input into the output module and output a final functional status grading result; the functional status grading result output by the output module includes: normal, mild functional impairment, moderate functional impairment, and severe functional impairment; the assessment and grading method includes: S1. Collect the original data of ICF-RS quantitative standard functional status classification, pre-process the original data and divide the data into training set, validation set and test set. The process of pre-processing the original data includes: S11. Eliminate data with missing values ​​or unreasonable data; S12, mapping the four labels of normal, mild dysfunction, moderate dysfunction, and severe dysfunction in the four dimensions to four preset numbers; S13, using the rule of selecting the majority, and using the evaluation results of multiple experts on a sample data as the label of the sample; S2. Construct neural network models corresponding to four different dimensions, use the training set to train the four neural network models respectively, use the loss function to calculate the loss value, use the back propagation technology to update the network parameters of the model, complete a round of training, and obtain the preliminary optimal model; the neural network models corresponding to the four different dimensions include: physical function dimension model, activity function dimension model, participation function dimension model, and overall function dimension model, among which, the data set used by the physical function dimension model includes 9 category data on physical function dimension in ICF, and physical function status assessment results; the data set used by the activity function dimension model includes 14 category data on daily activity dimension in ICF, and activity function status assessment results; the data set used by the participation function dimension model includes 7 category data on social participation dimension in ICF, and participation function status assessment results; the data set used by the overall function dimension model includes 30 category data in ICF-RS and overall function status assessment results; finally, the data sets are divided into training set, validation set and test set according to the set ratio; S3. Input the validation set into the preliminary optimal model to obtain a preliminary prediction result, determine whether the preliminary prediction result is consistent with the label value, calculate the accuracy of the current model on the validation set, and determine whether the accuracy is greater than the set value and is the highest. If so, save the model as the current optimal model, otherwise repeat the training process of step S2; S4. Test the current optimal model using the test set to obtain the function level evaluation performance result of the current optimal model on the test set, and determine whether the result is consistent with the true label value. If so, determine the current optimal model, otherwise repeat step S3; S5. Collect the patient's ICF-RS quantitative standard functional status data, input it into the current optimal model after preprocessing, and output the patient's functional level assessment in four dimensions.

2. An intelligent ICF-RS function evaluation and grading method according to claim 1, characterized in that: The ICF-RS quantitative standard functional status grading raw data described in step S1 includes the evaluation results of multiple experts, specifically: 30 ICF category evaluation results, physical function status evaluation results, activity function status evaluation results, participation function status evaluation results, and overall function status evaluation results.

3. An intelligent ICF-RS function evaluation and grading method according to claim 1, characterized in that: The process of constructing the neural network models corresponding to the four different dimensions is specifically as follows: the grid search method is used to determine the number of neuron nodes and the number of network layers of the four neural network models, wherein the overall function dimension model inputs 30 category data in ICF-RS, and outputs the overall function level assessment result; the body function dimension model inputs 9 category data of body function dimensions in ICF, and outputs the body function level assessment result; the activity function dimension model inputs 14 category data of daily activity dimensions in ICF, and outputs the activity function level assessment result; the participation function dimension model inputs 7 category data of social participation dimensions in ICF, and outputs the participation function level assessment result.

4. An intelligent ICF-RS function evaluation and grading method according to claim 3, characterized in that: The loss functions of the neural network models corresponding to the four different dimensions are all calculated using cross-entropy loss, and the stochastic gradient optimizer SGD is used to update and calculate the network parameters that affect the model training and output, so that the network parameters approach the optimal value, thereby minimizing the loss function.

5. An intelligent ICF-RS function evaluation and grading method according to claim 4, characterized in that: The training process described in step S2 is as follows: the training set of the corresponding dimensional model is used as the input data of the dimensional model, the output result is expressed as the predicted probability of the function level result, the output result is compared with the label value to calculate the loss value, and the stochastic gradient optimizer SGD is used to update and calculate the network parameters that affect the model training and output, so that the network parameters are close to the optimal value, thereby minimizing the loss function and obtaining a preliminary optimal model; The verification process in step S3 is as follows: the verification set of the corresponding dimensional model is used as the input data of the dimensional model, the accuracy of the output result compared with the label value is calculated, and when the accuracy is higher than the set value, the model is saved as the current optimal model; Among them, the iterative process of the neural network is realized by continuously repeating steps S2 and S3, thereby improving the functional level result evaluation performance of the neural network model.

6. An intelligent ICF-RS function evaluation and grading method according to any one of claims 1 to 5, characterized in that: The collected ICF-RS data set can continue to increase. After assisting medical staff to complete the four-dimensional level assessment of patients, the four-dimensional functional level assessment results obtained can be used as a new data set to conduct a new round of training for the saved current optimal model.

Images