Medical image key point capturing and evaluating method and device

Abstract

The invention discloses a medical image key point capturing and evaluating method and device, and relates to the technical field of medical detection, and the method comprises the steps: carrying out the key point capturing and classification of a to-be-detected human body through a convolutional neural network visual classifier; marking the coordinates of the captured key points of the human body, performing joint angle calculation according to the coordinates, and adding labels; inputting the labeled joint angle as a data sample into an adaptive neural fuzzy inference system so as to output a prediction result of the joint angle; and performing evaluation according to a prediction result and a set joint angle scoring rule. According to the method and the device, the problem that the evaluation precision is influenced by different postures due to low sensitivity of the input variable can be solved.

Description

Technical Field

[0001] The present application relates to the field of medical detection technology, and in particular to a method and device for capturing and evaluating key points of medical images. Background Art

[0002] In recent years, medical image key point detection has become increasingly common in the field of medical health. By detecting and analyzing the key points of the human body, it can play a good role in preventing and treating health problems caused by musculoskeletal diseases.

[0003] However, the existing key point detection system has low sensitivity to input variables, resulting in different postures affecting the evaluation accuracy. In addition, the insufficient precision of the input variables prevents researchers from accurately analyzing bone diseases. Moreover, additional post-processing is required to reduce the quantization error of data input, which increases the workload and reduces the speed of image analysis. Summary of the invention

[0004] The present application provides a method and device for evaluating key point capture in medical images, which can solve the problem that the sensitivity of input variables is low, resulting in different postures affecting the evaluation accuracy.

[0005] In a first aspect, an embodiment of the present application provides a medical image key point capture and evaluation method, the medical image key point capture and evaluation method comprising:

[0006] The key points of the human body to be detected are captured and classified through the convolutional neural network visual classifier;

[0007] Mark the coordinates of the captured key points of the human body, calculate the joint angles based on the coordinates, and add labels;

[0008] The labeled joint angles are input as data samples into the adaptive neuro-fuzzy inference system to output the prediction results of the joint angles;

[0009] Evaluation is performed based on the prediction results and the set joint angle scoring rules.

[0010] In combination with the first aspect, in one implementation, marking the captured coordinates of key points of the human body and calculating the joint angles according to the coordinates includes:

[0011] Calculate the three-dimensional coordinates of key points of the human body;

[0012] According to the classification of key points of the human body, the Euclidean distance and the cosine of the two-dimensional vector angle are calculated based on the three-dimensional coordinates to obtain the corresponding joint angle.

[0013] In combination with the first aspect, in one embodiment, the step of inputting the labeled joint angle as a data sample into an adaptive neural fuzzy inference system to output a prediction result of the joint angle includes:

[0014] Divide the data samples into training set, validation set, and test set and determine the data ratio;

[0015] Formulate the initial parameters of the adaptive neuro-fuzzy inference system and run it, and modify the parameter values ​​to obtain the best parameters through sensitivity analysis to obtain the best model of the adaptive neuro-fuzzy inference system;

[0016] Calculations are performed based on the best model to output predictions of joint angles.

[0017] In combination with the first aspect, in one implementation, modifying the parameter value through sensitivity analysis to obtain the optimal parameter includes:

[0018] The adaptive neural fuzzy inference system selects the root mean square error as an indicator, and judges the model accuracy by comparing the indicator performance under different parameters to determine the optimal parameters.

[0019] In combination with the first aspect, in one implementation, the joint angle scoring rule is provided with load levels of four joint angles and scores corresponding to the load levels.

[0020] In a second aspect, an embodiment of the present application provides a medical image key point capture evaluation device, the medical image key point capture evaluation device comprising a key point capture model and a fuzzy reasoning evaluation model;

[0021] The key point capture model includes a convolutional neural network visual classifier and a key point positioning set, and the fuzzy inference evaluation model includes an adaptive neural fuzzy inference system and a key point evaluation system;

[0022] The convolutional neural network visual classifier is used to capture and classify key points of the human body to be detected;

[0023] The key point positioning set is used to mark the coordinates of the captured human body key points, calculate the joint angles according to the coordinates, and add labels;

[0024] The adaptive neuro-fuzzy inference system is used to calculate the labeled joint angle as a data sample to output a prediction result of the joint angle;

[0025] The key point evaluation system is used to perform evaluation based on the prediction results and the set joint angle scoring rules.

[0026] In conjunction with the second aspect, in one implementation, the key point positioning set marks the coordinates of the captured human key points, and calculates the joint angles according to the coordinates, including:

[0027] Calculate the three-dimensional coordinates of key points of the human body;

[0028] According to the classification of key points of the human body, the Euclidean distance and the cosine of the two-dimensional vector angle are calculated based on the three-dimensional coordinates to obtain the corresponding joint angle.

[0029] In conjunction with the second aspect, in one implementation, the adaptive neural fuzzy inference system uses the labeled joint angle as a data sample to calculate and output a prediction result of the joint angle, including:

[0030] Divide the data samples into training set, validation set, and test set and determine the data ratio;

[0031] Formulate the initial parameters of the adaptive neuro-fuzzy inference system and run it, and modify the parameter values ​​to obtain the best parameters through sensitivity analysis to obtain the best model of the adaptive neuro-fuzzy inference system;

[0032] Calculations are performed based on the best model to output predictions of joint angles.

[0033] In conjunction with the second aspect, in one implementation, modifying the parameter value through sensitivity analysis to obtain the optimal parameter includes:

[0034] The adaptive neural fuzzy inference system selects the root mean square error as an indicator, and judges the model accuracy by comparing the indicator performance under different parameters to determine the optimal parameters.

[0035] In combination with the second aspect, in one implementation, the joint angle scoring rule is provided with load levels of four joint angles and scores corresponding to the load levels.

[0036] The beneficial effects brought by the technical solution provided by the embodiment of the present application include at least:

[0037] The medical image key point capture and evaluation method in the present application uses a convolutional neural network visual classifier to capture and classify key points of a human body to be detected; marks the coordinates of the captured human key points, calculates the joint angles based on the coordinates, and adds labels; inputs the labeled joint angles as data samples into an adaptive neural fuzzy inference system to output the prediction results of the joint angles; and performs evaluation based on the prediction results and the set joint angle scoring rules.

[0038] When the present application conducts risk assessment, its fuzzy logic theory can produce a gradual and relatively stable scoring result for inaccurate changes in joint angles, and perform fuzzy optimization on the basis of the original score to reduce the variance in the assessment results, thereby avoiding sudden changes in the assessment score and improving the reliability of the scoring result.

[0039] In addition, based on the initial fuzzy laws, the adaptive neural fuzzy reasoning system not only meets the changing characteristics of the fuzzy laws, but also maintains the learning mechanism and system of the artificial neural network. Through the continuous learning of the neural network, it adaptively adjusts the variables, membership functions and image parameters to meet the input and output association of the fuzzy reasoning system, and finally achieves the optimal setting of the network system parameters. The fuzzy model of the adaptive neural fuzzy reasoning system is a nonlinear fuzzy control method that combines the advantages of fuzzy logic and neural networks. It can adjust the condition parameters and result parameters through a hybrid learning algorithm, thereby automatically generating fuzzy laws and converting fuzzy values ​​into more specific and precise numerical values. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic diagram of a flow chart of an embodiment of a method for capturing and evaluating key points of medical images of the present application;

[0041] Figure 2 This is a structural block diagram of an embodiment of a medical image key point capture and evaluation device of the present application.

[0042] Figure 3 A flowchart that captures the key points of this application;

[0043] Figure 4 This is the structural hierarchy diagram of the adaptive neuro-fuzzy inference system of this application;

[0044] Figure 5 This is a score prediction flowchart for the fuzzy reasoning evaluation model of this application;

[0045] Figure 6 This is a graph showing the angle scores of the upper arm deviating from the center position in the embodiment of the present application;

[0046] Figure 7 This is a graph showing the angle scores of the lower arm deviating from the center position in the embodiment of the present application;

[0047] Figure 8 This is a score diagram of the angle of the wrist deviating from the center position in the embodiment of the present application;

[0048] Fig. 9 The angle of the neck deviating from the center position in the embodiment of the present application;

[0049] Fig.10 It is the angle of the torso deviating from the center position in the embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0051] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit "first", "second" and "third" to different types.

[0052] In the description of the embodiments of the present application, "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a specific way.

[0053] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0054] In some processes described in the embodiments of the present application, multiple operations or steps that appear in a specific order are included, but it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or in parallel, and the sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0055] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0056] In a first aspect, an embodiment of the present application provides a method for capturing and evaluating key points of medical images.

[0057] In one embodiment, referring to Figure 1 , Figure 1 This is a flow chart of an embodiment of the medical image key point capture and evaluation method of the present application. Figure 1 As shown, the evaluation method for capturing key points of medical images includes:

[0058] S1, using the convolutional neural network visual classifier to capture and classify key points of the human body to be detected;

[0059] The structure of the convolutional neural network visual classifier can be roughly divided into two parts: a backbone network (backbone) for extracting certain high-level features of the image and a classifier (classifier) ​​for completing the target category classification. In this embodiment, the human body image is mainly extracted and then divided into the facial layer and the limb layer according to the category.

[0060] S2, mark the coordinates of the captured human body key points, calculate the joint angles according to the coordinates, and add labels;

[0061] Specifically, in step S2, the three-dimensional coordinates of the key points of the human body are calculated; then, according to the classification of the key points of the human body, the Euclidean distance and the cosine of the angle between the two-dimensional vectors are calculated based on the three-dimensional coordinates to obtain the corresponding joint angle.

[0062] S3, inputting the labeled joint angles as data samples into the adaptive neuro-fuzzy inference system to output the prediction results of the joint angles;

[0063] Specifically, step S3 includes:

[0064] S31, dividing the data samples into a training set, a validation set, and a test set and determining the data ratio;

[0065] S32, formulating initial parameters of the adaptive neuro-fuzzy inference system and running it, modifying the parameter values ​​through sensitivity analysis to obtain optimal parameters, so as to obtain the optimal model of the adaptive neuro-fuzzy inference system;

[0066] Specifically, in this embodiment, the adaptive neural fuzzy inference system selects the root mean square error as an indicator, and judges the model accuracy by comparing the indicator performance under different parameters to determine the optimal parameters.

[0067] S33. Perform calculation based on the optimal model to output the prediction result of the joint angle.

[0068] S4. Evaluate based on the prediction results and the set joint angle scoring rules.

[0069] In this embodiment, the joint angle scoring rule is provided with four load levels of joint angles and scores corresponding to the load levels. After the prediction result of the joint angle is obtained, the joint angle scoring rule can be used for scoring.

[0070] It is worth noting that the above steps can be implemented by Figure 2 The medical image key point capture and evaluation device shown is implemented.

[0071] Among them, the medical image key point capture and evaluation device includes a key point capture model and a fuzzy reasoning evaluation model; the key point capture model includes a convolutional neural network visual classifier and a key point positioning set, and the fuzzy reasoning evaluation model includes an adaptive neural fuzzy reasoning system and a key point evaluation system.

[0072] The convolutional neural network visual classifier is used to capture and classify key points of the human body to be detected; the key point positioning set is used to mark the coordinates of the captured human body key points, calculate the joint angles according to the coordinates, and add labels.

[0073] The adaptive neuro-fuzzy inference system is used to calculate the labeled joint angles as data samples to output the prediction results of the joint angles; the key point evaluation system is used to evaluate according to the prediction results and the set joint angle scoring rules.

[0074] The above steps are further explained below in conjunction with a medical image key point capture and evaluation device:

[0075] The key point capture model captures the human body to be detected through a convolutional neural network visual classifier, and locates the key points of the whole body through a key point positioning set. The positioning principle follows the top-down principle.

[0076] See also Figure 3 As shown, the key point capture model includes a functional layer, a main layer, a facial layer and a limb layer. The functional layer is composed of two convolutional layers and a bottleneck block. The main layer is a multi-resolution network composed of 38 output channels. The main layer predicts the body and foot key points and the face and hand bounding boxes through the features extracted from the functional layer. Each bounding box is represented by four corner points and a midpoint. The facial layer uses the facial bounding box predicted by the main layer, and the limb layer uses the limb bounding box predicted by the main layer.

[0077] Specifically, the functional layer processes the input image and extracts shared features (Z1 and Z2). When the input size is 384px×288px image, the output feature map sizes of Z1 and Z2 are 192px×144px and 96px×72px images respectively.

[0078] The main layer is a multi-resolution network with 38 output channels, which generates 38 key points for each human body at the same time. The main layer predicts the key points of the body and feet, as well as the bounding boxes of the face and hands through the features extracted from the functional layer. Each bounding box is represented by four corner points and a center point, which means that each bounding box consists of five points, where the four corner points represent the four vertices of the bounding box and the center point represents the center position of the bounding box. These points can be used to describe the shape, size, and position of the bounding box, thereby helping computer vision algorithms to identify and locate objects in images. For example, in object detection tasks, each bounding box can be used to represent the position and shape of an object, thereby helping the algorithm to identify and locate objects in images.

[0079] The facial layer and limb layer use the facial and hand bounding boxes predicted by the main layer, and crop features from the corresponding areas of Z1 and Z2. The feature size in Z1 is adjusted to 64px×64px image, and the feature size in Z2 is adjusted to 32px×32px image. Based on the facial layer and limb layer, the heat map of the key points of the face and hand is predicted simultaneously.

[0080] The key point positioning set includes human body region, face region, left region and right region and 133 key points, specifically, 17 body key points, 6 foot key points, 68 face key points, 42 hand key points. The face and hand regions are the minimum bounding rectangles for key point detection. When the face and hand image clarity meets the standard, the face and hand regions are marked as valid and labels are added to the key points. Otherwise, when blur or severe occlusion occurs, the key points are considered invalid. Among them, the 17 body key points are as follows: nose, right eye, left eye (LI), right shoulder, left shoulder, right elbow, left elbow, right wrist, left wrist, right torso, left torso, right knee, left knee (LK), right ankle, left ankle, right joint and left joint.

[0081] The key point capture model identifies the key points of the human body, calculates the three-dimensional position of the key points of the human body, establishes a coordinate system, and establishes the relationship between the center of the human joint and its local coordinate system based on the coordinate system, and uses the Euclidean distance and the cosine of the two-dimensional vector angle to obtain the joint angle.

[0082] It is worth noting that in the commonly used evaluation methods, when the joint angle is in the boundary range, the subjective determination of the traditional method is less sensitive to the input variables. Different postures may result in the same risk score. However, this embodiment uses a key point capture model to identify the key points of the human body and calculate their three-dimensional positions (x, y, z), and then calculate the joint angles, which can avoid the above problems well.

[0083] See also Figure 4 As shown, the adaptive neuro-fuzzy reasoning system includes a 5-layer reasoning architecture, two input variables and two fuzzy rules. The reasoning architecture includes an input layer, a rule layer, a normalization layer, a conclusion layer, and a defuzzification layer.

[0084] Specifically, the input layer selects sample data and inputs it, and outputs the membership degree of each input through fuzzy rules, which is in the form of:

[0085]

[0086] Where: They represent the fuzzy set membership functions based on fuzzy rules in the input layer; μA i ,μB j The membership function of each data set input is represented by the triangular membership function, and the calculation formula is as follows:

[0087]

[0088] Where: Where (i=1,2), Among them, (j=1,2) the second layer are all premise functions.

[0089] The rule layer uses fuzzy rules to count the incentive strength of each rule and output the result. The formula is:

[0090]

[0091] Where: is the comprehensive applicability of each rule, ω i is the incentive strength of the i-th rule.

[0092] The normalization layer supervises the excitation process of each rule, and its input comes from the output of the rule layer. For example, the normalized applicability of the i-th rule can be determined as follows:

[0093]

[0094] Where: is the normalized applicability of each rule, is the applicability ratio of the ith rule based on the total number of rules.

[0095] The conclusion layer is used to calculate the contribution of each rule and output each adaptive node. The expression is:

[0096]

[0097] Where: is the input value before defuzzification.

[0098] The defuzzification layer calculates the sum of input and output in the fuzzy operation system under each rule. The calculation formula is as follows:

[0099]

[0100] Where: Q 5 is the sum of the input and output in the system.

[0101] The adaptive neuro-fuzzy inference system is a nonlinear fuzzy control method, which includes two input variables and two fuzzy rules. Its expression is as follows:

[0102] Rule 1: y 1 =A 1 ,y 2 =B 1 ,f 1 =p 1 y 1 +q 1 y 1 +r 1 ;

[0103] Rule 2: y 1 =A 2 ,y 2 =B 2 ,f 2 =p 2 y 1 +q 2 y 1 +r 2 ;

[0104] Among them, y 1 ,y 2 There are two input variables, A i , B i (i=1,2) is y 1 ,y 2 The fuzzy sum of i ,q i 、r i (i=1,2) is the conclusion parameter of the adaptive neuro-fuzzy inference system, f i (i=1,2) is the linear combination form representing the fuzzy rules.

[0105] Furthermore, in this embodiment, the adaptive neural fuzzy inference system selects the root mean square error as an indicator, compares the differences in the model prediction results under the input data sample parameters, and selects the optimal parameters to optimize the accuracy of the final prediction results. The indicator calculation formula is as follows:

[0106]

[0107] in, is the predicted value, y t is the actual value, and RMSE is the result accuracy.

[0108] See also Figure 5 As shown, it is the score prediction process of the fuzzy reasoning evaluation model. It is worth mentioning that the model has the characteristics of multi-layer feedforward network technology. The weights in each layer can be adaptively adjusted according to the given sample data results, so that the model can better reflect the characteristics of the actual system.

[0109] Fuzzy logic does not completely use 1 or 0 to form an exact value. When using a value between 0 and 1, it can be expressed as an intermediate value. Since there are decision-making problems with subjective uncertainty in the field of human factors engineering risk assessment, when applying the key point assessment system for risk assessment, fuzzy logic theory can produce a gradual transition and relatively stable scoring result for inaccurate joint angle changes, and perform fuzzy optimization on the basis of the original score to reduce the difference in the evaluation result, which can avoid sudden changes in the evaluation score and improve the reliability of the scoring result.

[0110] On the basis of the initial fuzzy laws, the adaptive neural fuzzy reasoning system not only meets the changing characteristics of the fuzzy laws, but also maintains the learning mechanism and system of the artificial neural network. Through the continuous learning of the neural network, it adaptively adjusts the variables, membership functions and image parameters to meet the input and output association of the fuzzy reasoning system, and finally achieves the optimal setting of the network system parameters. The fuzzy model of the adaptive neural fuzzy reasoning system is a nonlinear fuzzy control method. It combines the advantages of fuzzy logic and neural networks. It can adjust the condition parameters and result parameters through a hybrid learning algorithm, thereby automatically generating fuzzy laws and converting fuzzy values ​​into more specific and precise numerical values.

[0111] The key point assessment system includes 4 joint angle load levels and corresponding scores. The first level corresponds to a score of 1-2, the second level corresponds to a score of 3-4, the third level corresponds to a score of 5-6, and the fourth level corresponds to a score of ≥7.

[0112] The critical point assessment system assessment process is divided into two parts, the first part is to assess the upper arm, lower arm and wrist (Group A), and the second part is to assess the neck, torso and legs (Group B). Figures 6 to 8The joint angle shown is the first part of the input. The initial score (Group A score) can be obtained by the joint angle load level and the score corresponding to the load level. Fig. 9 and Fig.10 Get the score of the second part (Group B), and get the evaluation result through the 4 different levels of scores corresponding to different results;

[0113] Level 1 is 1-2 points. In level 1, if the human body posture is not required or repeated for a long time, there is no need to adjust the posture;

[0114] Level 2 is 3-4 points. In level 2, it indicates that the human body posture is uncomfortable, or even if it is in the comfortable range, it requires repetitive movements and has static loads or requires force, and the operation needs to be analyzed and adjusted;

[0115] Level 3 is 5-6 points. In level 3, it indicates that the human body posture is uncomfortable, repetitive movements are required, there is a static load or force is required, and the operation needs to be analyzed and adjusted as soon as possible;

[0116] Level 4 is ≥7 points. In the Level 4 certificate, it indicates that the working posture is at or close to the limit of activity, requires repetitive movements and is overloaded, and the posture needs to be adjusted immediately to reduce the load;

[0117] In the process of human key point detection, the face and hands have high requirements for positioning accuracy. At the same time, the scales of the body, face and hands vary greatly. This system relies on key point positioning to add additional key point annotations of the face, hands and feet to form a complete full-body annotation, which improves the credibility of the evaluation results.

[0118] The medical image key point capture and evaluation method in the present application uses a convolutional neural network visual classifier to capture and classify key points of a human body to be detected; marks the coordinates of the captured human key points, calculates the joint angles based on the coordinates, and adds labels; inputs the labeled joint angles as data samples into an adaptive neural fuzzy inference system to output the prediction results of the joint angles; and performs evaluation based on the prediction results and the set joint angle scoring rules.

[0119] When the present application conducts risk assessment, its fuzzy logic theory can produce a gradual and relatively stable scoring result for inaccurate changes in joint angles, and perform fuzzy optimization on the basis of the original score to reduce the variance in the assessment results, thereby avoiding sudden changes in the assessment score and improving the reliability of the scoring result.

[0120] In addition, based on the initial fuzzy laws, the adaptive neural fuzzy reasoning system not only meets the changing characteristics of the fuzzy laws, but also maintains the learning mechanism and system of the artificial neural network. Through the continuous learning of the neural network, it adaptively adjusts the variables, membership functions and image parameters to meet the input and output association of the fuzzy reasoning system, and finally achieves the optimal setting of the network system parameters. The fuzzy model of the adaptive neural fuzzy reasoning system is a nonlinear fuzzy control method that combines the advantages of fuzzy logic and neural networks. It can adjust the condition parameters and result parameters through a hybrid learning algorithm, thereby automatically generating fuzzy laws and converting fuzzy values ​​into more specific and precise numerical values.

[0121] In a second aspect, an embodiment of the present application also provides a medical image key point capture and evaluation device.

[0122] In one embodiment, referring to Figure 2 , Figure 2 This is a functional module diagram of an embodiment of a medical image key point capture and evaluation device of the present application. Figure 2 As shown, the medical image key point capture evaluation device includes a key point capture model and a fuzzy reasoning evaluation model.

[0123] Wherein, the key point capture model includes a convolutional neural network visual classifier and a key point positioning set, and the fuzzy inference evaluation model includes an adaptive neural fuzzy inference system and a key point evaluation system;

[0124] The convolutional neural network visual classifier is used to capture and classify key points of the human body to be detected;

[0125] The key point positioning set is used to mark the coordinates of the captured human body key points, calculate the joint angles according to the coordinates, and add labels;

[0126] The adaptive neuro-fuzzy inference system is used to calculate the labeled joint angle as a data sample to output a prediction result of the joint angle;

[0127] The key point evaluation system is used to perform evaluation based on the prediction results and the set joint angle scoring rules.

[0128] Furthermore, in one embodiment, the key point positioning set marks the coordinates of the captured human key points, and calculates the joint angles according to the coordinates, including:

[0129] Calculate the three-dimensional coordinates of key points of the human body;

[0130] According to the classification of key points of the human body, the Euclidean distance and the cosine of the two-dimensional vector angle are calculated based on the three-dimensional coordinates to obtain the corresponding joint angle.

[0131] Furthermore, in one embodiment, the adaptive neural fuzzy inference system uses the labeled joint angle as a data sample to calculate and output a prediction result of the joint angle, including:

[0132] Divide the data samples into training set, validation set, and test set and determine the data ratio;

[0133] Formulate the initial parameters of the adaptive neuro-fuzzy inference system and run it, and modify the parameter values ​​to obtain the best parameters through sensitivity analysis to obtain the best model of the adaptive neuro-fuzzy inference system;

[0134] Calculations are performed based on the best model to output predictions of joint angles.

[0135] Furthermore, in one embodiment, modifying the parameter values ​​through sensitivity analysis to obtain the optimal parameters includes:

[0136] The adaptive neural fuzzy inference system selects the root mean square error as an indicator, and judges the model accuracy by comparing the indicator performance under different parameters to determine the optimal parameters.

[0137] Furthermore, in one embodiment, the joint angle scoring rule is provided with load levels of four joint angles and scores corresponding to the load levels.

[0138] Among them, the functional implementation of each module in the above-mentioned medical image key point capture and evaluation device corresponds to the various steps in the above-mentioned medical image key point capture and evaluation method embodiment, and its functions and implementation processes will not be repeated here one by one.

[0139] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for capturing and evaluating key points of medical images, characterized in that: The medical image key point capture and evaluation method comprises: The key points of the human body to be detected are captured and classified through the convolutional neural network visual classifier; Mark the coordinates of the captured key points of the human body, calculate the joint angles based on the coordinates, and add labels; The labeled joint angles are input as data samples into the adaptive neuro-fuzzy inference system to output the prediction results of the joint angles; Evaluation is performed based on the prediction results and the set joint angle scoring rules.

2. The medical image key point capture and evaluation method according to claim 1, characterized in that: The captured coordinates of the key points of the human body are marked, and the joint angles are calculated according to the coordinates, including: Calculate the three-dimensional coordinates of key points of the human body; According to the classification of key points of the human body, the Euclidean distance and the cosine of the two-dimensional vector angle are calculated based on the three-dimensional coordinates to obtain the corresponding joint angle.

3. The medical image key point capture and evaluation method according to claim 1, characterized in that: The labeled joint angles are input as data samples into the adaptive neural fuzzy inference system to output the prediction results of the joint angles, including: Divide the data samples into training set, validation set, and test set and determine the data ratio; Formulate the initial parameters of the adaptive neuro-fuzzy inference system and run it, and modify the parameter values ​​to obtain the best parameters through sensitivity analysis to obtain the best model of the adaptive neuro-fuzzy inference system; Calculations are performed based on the best model to output predictions of joint angles.

4. The medical image key point capture and evaluation method according to claim 3, characterized in that: The parameter values ​​are modified through sensitivity analysis to obtain the optimal parameters, including: The adaptive neural fuzzy inference system selects the root mean square error as an indicator, and judges the model accuracy by comparing the indicator performance under different parameters to determine the optimal parameters.

5. The medical image key point capture and evaluation method according to claim 1, characterized in that: The joint angle scoring rule is provided with load levels of four joint angles and scores corresponding to the load levels.

6. A medical image key point capture and evaluation device, characterized in that: The medical image key point capture and evaluation device comprises a key point capture model and a fuzzy reasoning evaluation model; The key point capture model includes a convolutional neural network visual classifier and a key point positioning set, and the fuzzy inference evaluation model includes an adaptive neural fuzzy inference system and a key point evaluation system; The convolutional neural network visual classifier is used to capture and classify key points of the human body to be detected; The key point positioning set is used to mark the coordinates of the captured human body key points, calculate the joint angles according to the coordinates, and add labels; The adaptive neuro-fuzzy inference system is used to calculate the labeled joint angle as a data sample to output a prediction result of the joint angle; The key point evaluation system is used to perform evaluation based on the prediction results and the set joint angle scoring rules.

7. The medical image key point capture and evaluation device according to claim 6, characterized in that: The key point positioning set marks the coordinates of the captured human key points and calculates the joint angles according to the coordinates, including: Calculate the three-dimensional coordinates of key points of the human body; According to the classification of key points of the human body, the Euclidean distance and the cosine of the two-dimensional vector angle are calculated based on the three-dimensional coordinates to obtain the corresponding joint angle.

8. The medical image key point capture and evaluation device according to claim 6, characterized in that: The adaptive neural fuzzy inference system uses the labeled joint angles as data samples to calculate and output the prediction results of the joint angles, including: Divide the data samples into training set, validation set, and test set and determine the data ratio; Formulate the initial parameters of the adaptive neuro-fuzzy inference system and run it, and modify the parameter values ​​to obtain the best parameters through sensitivity analysis to obtain the best model of the adaptive neuro-fuzzy inference system; Calculations are performed based on the best model to output predictions of joint angles.

9. The medical image key point capture and evaluation device according to claim 8, characterized in that: The parameter values ​​are modified through sensitivity analysis to obtain the optimal parameters, including: The adaptive neural fuzzy inference system selects the root mean square error as an indicator, and judges the model accuracy by comparing the indicator performance under different parameters to determine the optimal parameters.

10. The medical image key point capture and evaluation device according to claim 6, characterized in that: The joint angle scoring rule is provided with load levels of four joint angles and scores corresponding to the load levels.

Images