Auxiliary diagnosis method and system for foot sports injuries based on gait data characteristics

Abstract

The present invention provides a method and system for auxiliary diagnosis of foot sports injuries based on gait data features, the method comprising: data preprocessing, gait data feature space construction, injury auxiliary diagnosis and verification of its effectiveness; the present invention solves the problem of shortage of intelligent and accurate diagnostic methods in the existing diagnosis and treatment of foot sports injuries; based on the collection of foot motion information of subjects, the range of motion of joints in the natural walking process is finely described based on the Heidelberg foot measurement method, and a gait data feature space of foot injuries is established to realize auxiliary diagnosis of foot sports injuries; the method and system provide a new and effective intelligent auxiliary tool for quantitative, comprehensive and convenient foot injury diagnosis, and at the same time, provide necessary anthropometric basis for research and application in related fields such as the development of foot rehabilitation assistive devices and the development of lower limb exoskeleton robots.

Description

Technical Field

[0001] The present invention relates to the technical field of human motion feature analysis and intelligent auxiliary diagnosis, and in particular to a method and system for auxiliary diagnosis of foot motion injuries based on gait data features. Background Art

[0002] Compared with the hip and knee joints, the foot has the smallest surface and bears almost all of the body's weight. As an important weight-bearing fulcrum for high-frequency movements such as walking, running, and jumping, the foot is undoubtedly the most vulnerable joint in the human body.

[0003] This situation highlights the limitations of traditional clinical diagnosis and treatment in terms of time, space, standardization, and accuracy. The increasing maturity of accurate detection technology for human kinematic parameters provides objective and effective data support for auxiliary diagnosis of foot sports injuries such as talocervical ligament rupture, and also provides the possibility for the construction of intelligent auxiliary diagnosis methods and systems for injuries.

[0004] At present, although the medical field has conducted extensive research on the pathogenesis, diagnosis, treatment and rehabilitation methods of foot and ankle sports injuries, it is limited by the "descriptive medicine" method dominated by clinical experience, and has not yet formed a quantitative and accurate "explanatory medicine" intelligent diagnostic model. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an auxiliary diagnosis method and system for foot sports injuries based on gait data characteristics, so as to solve the problem of shortage of intelligent and accurate diagnosis methods in the existing diagnosis and treatment of foot sports injuries. The purpose is to describe the gait data characteristics of the foot sports injury group and the control group with labeled data, and to construct an auxiliary diagnosis method and system for foot sports injuries, so as to provide a new and effective intelligent auxiliary tool for quantitative, comprehensive and convenient foot injury diagnosis. At the same time, it also provides the necessary anthropometric basis for the research and application in related fields such as the development of foot rehabilitation assistive devices and the development of lower limb exoskeleton robots.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] A method for auxiliary diagnosis of foot sports injuries based on gait data features, the method comprising:

[0008] Collect the gait motion information of the subjects during natural walking for preprocessing;

[0009] Based on the preprocessed gait motion information, the gait data features of the injury group and the control group in 8 dimensions were calculated to construct the gait data feature space;

[0010] A spatiotemporal Transformer network was constructed, and the gait data features of the injury group and the control group were used to form a training set to train the network to achieve auxiliary diagnosis of foot sports injuries.

[0011] Furthermore, after achieving auxiliary diagnosis of foot sports injuries, the method further includes:

[0012] Construct a confusion matrix to record the number of true positives, false positives, true negatives, and false negatives of injury-assisted diagnosis results;

[0013] The sensitivity and specificity of the auxiliary diagnosis method for foot sports injuries were calculated to form a receiver operating characteristic curve, and the effectiveness of the decision-making method was verified by the area under the receiver operating characteristic curve.

[0014] Furthermore, the collecting of the subject's gait motion information includes:

[0015] Based on the Heidelberg foot measurement method, a foot and ankle motion measurement model covering the knee joint, foot part, and foot is established; based on the foot and ankle motion measurement model, the gait data of the subject is collected through an optical motion capture system.

[0016] Furthermore, the preprocessing of the collected gait motion information includes:

[0017] Perform low-pass digital filtering on the collected gait motion information to reduce the noise of the original motion information;

[0018] According to the common characteristics of human gait and the actual movement of the subjects, the filtered motion information is automatically segmented into gait cycles according to the spatial coordinates of the dorsal calcaneal markers.

[0019] An interpolation model is constructed to standardize the number of sampling points in the gait cycle.

[0020] Furthermore, the constructing of the gait data feature space includes:

[0021] Gait data features of foot sports injuries are extracted, and the extracted gait data features include: tibia-talar flexion angle, forefoot-heel abduction angle, medial arch angle of the foot, lateral arch angle of the foot, rotation angle, forefoot-ankle external rotation angle, medial-lateral foot mid-spin angle, and ankle-heel extension and retraction distance.

[0022] Furthermore, the method of implementing auxiliary diagnosis of foot sports injuries includes:

[0023] The gait data features of the injury group and the control group were integrated to form a training set for the auxiliary diagnosis model of foot sports injuries;

[0024] Construct a foot sports injury auxiliary diagnosis method based on the spatiotemporal Transformer network, input the training set into the network, and learn the mapping relationship from gait data features to injury decisions;

[0025] Optimize the decision network parameters and iterate repeatedly until the output is a discrimination result that meets the loss function requirements and accuracy requirements.

[0026] Accordingly, in order to solve the above technical problems, the present invention also provides the following technical solutions:

[0027] A foot sports injury auxiliary diagnosis system based on gait data features, the system comprising:

[0028] A data preprocessing module, which is used to filter the raw data of the collected gait motion information, automatically segment the gait cycle of the filtered motion information according to the spatial coordinate positions of the feature markers based on the common characteristics of human gait and the actual motion conditions of the individual subjects, and perform gait cycle standardization processing;

[0029] A gait data feature space construction module, wherein the gait data feature space construction module is used to calculate the gait data features of the injury group and the control group in 8 dimensions according to the gait cycle segmentation and standardization results, construct a gait data feature space, and at the same time, construct a gait data feature relationship matrix according to the spatial relationship between the gait data features in 8 dimensions and the marking points;

[0030] The injury auxiliary diagnosis module is used to construct a spatiotemporal Transformer network, and the network is trained using a training set composed of gait data features of the injury group and the control group to realize the auxiliary diagnosis function of foot sports injuries.

[0031] The method and system validity verification module are used to construct a confusion matrix, calculate the sensitivity and specificity of the auxiliary diagnosis model for foot sports injuries, and verify the validity of the decision-making method by verifying the area under the receiver operating characteristic curve to verify the validity of the decision-making model.

[0032] The beneficial effects of the above technical solution of the present invention are as follows:

[0033] In response to the problem of shortage of intelligent and accurate diagnostic methods in the existing diagnosis and treatment of foot sports injuries, the present invention provides an auxiliary diagnosis method and system for foot sports injuries based on gait data features. The foot and ankle motion information involved in the present invention comes from an optical motion capture system based on an infrared camera. On this basis, after data preprocessing, a gait data feature space is constructed for the injury group and the control group. Using the constructed gait data features as a training set, an auxiliary diagnosis method and system for foot sports injuries and a method for verifying their effectiveness are developed, thereby providing a high-quality and efficient intelligent auxiliary tool for the diagnosis of foot sports injuries.

[0034] The foot sports injury auxiliary diagnosis method and system based on gait data features of the present invention can effectively improve the level of accurate description of foot and ankle gait data features, realize quantitative and accurate foot sports injury decision-making, and provide an intelligent auxiliary diagnosis tool for foot sports injuries. It is widely used in multiple related aspects such as acute diagnosis of foot sports injuries, rehabilitation evaluation, and development of intelligent assistive devices. By constructing a quantitative and accurate "explanatory medicine" diagnostic model, algorithm and system, it will help the development of clinical medicine in related fields in my country.

[0035] The foot sports injury auxiliary diagnosis method and system based on gait data features of the present invention, combined with advanced intelligent science and technology, introduces new interdisciplinary thinking into the domestic sports medicine and rehabilitation industry, helps to explore and establish a complete intelligent service theory and algorithm for sports injury patients under the "medical-engineering integration" model, and helps to improve the standardization, process engineering application and innovation construction level in my country's medical equipment, health interaction equipment and other fields. It has important theoretical significance for strengthening the new technologies, new formats and new models of "new generation artificial intelligence" in the medical field. At the same time, it has a promoting effect on the construction of intelligent medical services in my country. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of the overall process of the method and system for auxiliary diagnosis of foot sports injuries based on gait data features of the present invention;

[0037] Figure 2 The overall network structure diagram of the spatiotemporal Transformer module in the foot sports injury auxiliary diagnosis method and system based on gait data features of the present invention;

[0038] Figure 3 It is a structural schematic diagram of a spatiotemporal module unit in a spatiotemporal Transformer module in a foot sports injury auxiliary diagnosis method and system based on gait data features of the present invention;

[0039] Figure 4A network structure diagram of a spatial Transformer module in a spatiotemporal Transformer module in a foot sports injury auxiliary diagnosis method and system based on gait data features of the present invention;

[0040] Figure 5 A network structure diagram of a temporal Transformer module in a spatiotemporal Transformer module in a foot sports injury auxiliary diagnosis method and system based on gait data features of the present invention;

[0041] Figure 6 This is a module diagram of the foot sports injury auxiliary diagnosis method and system based on gait data features of the present invention. DETAILED DESCRIPTION

[0042] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0043] First embodiment

[0044] This embodiment provides a method for auxiliary diagnosis of foot sports injuries based on gait data features, and the method for auxiliary diagnosis of foot sports injuries based on gait data features includes:

[0045] S101, collecting gait motion information of the subject;

[0046] It should be noted that, in this embodiment, the above S101 is specifically:

[0047] Based on the Heidelberg foot measurement method, a foot and ankle motion measurement model covering 12 bony landmarks in three parts, namely the knee joint, ankle joint and foot, was established; based on the foot and ankle motion measurement model, the gait motion information of the subjects was collected through an optical motion capture system based on an infrared camera.

[0048] S102, preprocessing the collected gait motion information;

[0049] It should be noted that, in this embodiment, the above S102 specifically includes the following process:

[0050] 1. Perform low-pass digital filtering on the collected gait data to reduce the noise caused by factors such as the collection system, environment, marker displacement, and occlusion;

[0051] 2. Based on the common characteristics of human gait and the actual movement of the individual subjects, the filtered motion information is automatically segmented into gait cycles according to the spatial coordinate positions of the dorsal calcaneal markers;

[0052] 3. The number of sampling points in each gait cycle is standardized by proportional interpolation.

[0053] S103, based on the preprocessed gait motion information, calculating the gait data features of the injury group and the control group in 8 dimensions, and constructing a gait data feature space and a gait data feature relationship matrix;

[0054] It should be noted that, in this embodiment, the above S103 is specifically:

[0055] 1. Gait data features of foot sports injuries are extracted, including: tibia-talar flexion angle, forefoot-heel abduction angle, medial arch angle of the foot, lateral arch angle of the foot, rotation angle, forefoot-ankle external rotation angle, medial-lateral foot mid-spin angle, ankle-heel extension distance.

[0056] 2. According to the spatial relationship between the gait data features in 8 dimensions and the 12 bony landmarks mentioned in S101, a feature relationship matrix is ​​constructed therebetween, and the matrix is ​​multiplied with its transposed matrix to obtain a relationship matrix of the gait data features in 8 dimensions, in which the diagonal elements are all set to 0.

[0057] S104, constructing a spatiotemporal Transformer network, using the gait data features of the injury group and the control group to form a training set to train the network, so as to achieve auxiliary diagnosis of foot sports injuries;

[0058] It should be noted that, in this embodiment, the above S105 specifically includes the following process:

[0059] 1. Integrate the gait data features of the injury group and the control group to form a training set for the auxiliary diagnosis model of foot sports injuries;

[0060] 2. Construct an auxiliary diagnosis method for foot sports injuries based on the spatiotemporal Transformer network, input the training set into the network, and learn the mapping relationship from gait data features to injury decisions;

[0061] 3. Optimize the decision network parameters and iterate repeatedly until the output is a judgment result that meets the requirements of the loss function, so as to achieve accurate auxiliary diagnosis of foot sports injuries and solve the problem of lack of intelligent and accurate diagnostic methods.

[0062] S105, calculate the area under the receiver operating curve to evaluate the effectiveness of the decision-making method and system.

[0063] It should be noted that, in this embodiment, the above S106 specifically includes the following process:

[0064] 1. Construct a confusion matrix to record the number of true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN) of injury-assisted diagnosis results;

[0065] 2. According to a series of binary cutoff values, calculate the sensitivity and specificity of the auxiliary diagnosis method for foot sports injuries, and draw the receiver operating characteristic (ROC) curve with sensitivity as the ordinate and specificity as the abscissa;

[0066] 3. Calculate the area under the receiver operating characteristic curve (AUC) to evaluate the effectiveness of the decision-making method.

[0067] In summary, this embodiment proposes an auxiliary diagnosis method for foot sports injuries based on gait data features, which mainly establishes a gait data feature space and a gait data feature relationship matrix with a basis for human motion measurement for the gait motion information collected by an optical motion capture system based on an infrared camera, constructs an auxiliary diagnosis model for foot sports injuries based on a spatiotemporal Transformer network, and verifies its effectiveness through the receiver operating characteristic curve, thereby providing an intelligent auxiliary tool for the diagnosis of foot sports injuries, thereby solving the problem of shortage of intelligent and accurate diagnostic methods in the existing diagnosis and treatment of foot sports injuries.

[0068] Second embodiment

[0069] See also Figure 1 This embodiment provides a method for auxiliary diagnosis of foot sports injuries based on gait data features. Figure 1 As shown, the following steps are included:

[0070] S101, constructing a preprocessing method for gait motion information;

[0071] It should be noted that, in this embodiment, the above S101 is specifically:

[0072] First, the Butterworth low-pass filter is used to digitally filter the original motion information of the foot and ankle. The specific filtering method is:

[0073] There is inevitably high-frequency noise in the original motion information of the foot and ankle collected by the optical motion capture system based on an infrared camera. Therefore, this method and system use a first-order zero-delay Butterworth low-pass filter for digital filtering with a sampling frequency of 100 Hz, a passband cutoff frequency of 0.01 Hz, and a stopband cutoff frequency of 20 Hz to smooth high-frequency noise and reduce interference information in the original data.

[0074] Secondly, an individual-specific automatic gait cycle segmentation method is established to segment the natural walking process according to the gait cycle. The specific segmentation method is as follows:

[0075] Heel strike is the main sign of the end of the previous gait cycle and the beginning of the current gait cycle. From the perspective of spatial motion characteristics, the dorsal side of the heel reaches an extremely low point in the entire gait motion at this moment. Therefore, this method uses the minimum value point of the CCL marker point located on the dorsal side of the calcaneus in the Z-axis direction in the fixed spatial coordinate system as the quantitative segmentation point to divide the gait cycle.

[0076] Then, a standardized processing method for gait cycle sampling points based on proportional interpolation is constructed. The specific standardized processing method is:

[0077] According to the gait cycle relative to the time axis, from the start (zero time) to the end, it is equally divided into 100 new sampling points. If the new sampling point coincides with the original sampling point in time, the value of the new sampling point is equal to the original value; if the new sampling point does not coincide with the original sampling point in time, the value of the new sampling point is:

[0078]

[0079] in, is the value of the new sampling point, is the time point of the new sampling point, and are respectively The two original sampling time points close to each other, and The original sampling time points and The corresponding value.

[0080] S102, constructing a gait data feature space and a gait data feature relationship matrix based on the Heidelberg foot measurement method;

[0081] It should be noted that, in this embodiment, the gait data features involved in the feature space of this embodiment include 8 dimensions: tibia-talar flexion angle, forefoot-heel abduction angle, medial arch angle of the foot, lateral arch angle of the foot, rotation angle, forefoot-ankle external rotation angle, medial-lateral foot mid-spin angle, ankle-heel extension distance:

[0082] (A) Tibial-talar flexion angle, specifically: the angle between the projection of the tibial axis on the plane perpendicular to the axis of rotation formed by the midpoint of the line connecting the medial and lateral malleolus landmarks (MML, LML) to the midpoint of the line connecting the medial and lateral epicondyle landmarks (MEP, LEP) and the projection of the talar axis on the plane perpendicular to the axis of rotation formed by the line connecting the P point to the navicular landmark (NAV), with the medial malleolus landmark (MML) to the lateral malleolus landmark (LML) as the axis of rotation, where: ;

[0083] (B) Forefoot-heel abduction angle, specifically: the angle between the projection of the forefoot axis from the distal end of the first metatarsal (DMT1) to the distal end of the fifth metatarsal (DMT5) on the plane perpendicular to the rotation axis, with the medial malleolus marker (MML) to the lateral malleolus marker (LML) as the rotation axis, and the projection of the calcaneal axis from the medial calcaneal marker (MCL) to the lateral calcaneal marker (LCL) on the plane perpendicular to the rotation axis;

[0084] (C) Medial arch angle, specifically: the angle between the medial foot axis from the navicular marker (NAV) to the distal metatarsal 1 (DMT1) and the medial foot axis from the navicular marker (NAV) to the medial calcaneal marker (MCL), where the rotation axis is the common perpendicular line between the medial foot axis and the medial foot axis;

[0085] (D) Lateral arch angle of the foot, specifically: the angle between the projection of the lateral axis of the front foot from the proximal end of the first metatarsal (PMT1) to the proximal end of the fifth metatarsal (PMT5) on the plane perpendicular to the rotation axis, and the projection of the lateral axis of the rear foot from the proximal end of the fifth metatarsal (PMT5) to the lateral end of the calcaneus (LCL) on the plane perpendicular to the rotation axis;

[0086] (E) Rotation angle, specifically: the angle between the projection of the foot axis from the medial malleolus landmark (MML) to the lateral malleolus landmark (LCL) on the plane perpendicular to the rotation axis and the projection of the heel axis from the medial calcaneal landmark (MCL) to the lateral calcaneal landmark (LCL) on the plane perpendicular to the rotation axis, with the rotation axis from point P to the navicular landmark (NAV);

[0087] (F) Forefoot-ankle external rotation angle, specifically: the angle between the projection of the foot axis from the medial malleolus marker (MML) to the lateral malleolus marker (LCL) on the plane perpendicular to the rotation axis and the projection of the forefoot axis from the distal end of the first metatarsal (DMT1) to the distal end of the fifth metatarsal (DMT5) on the plane perpendicular to the rotation axis, with the rotation axis from point P to the navicular marker (NAV);

[0088] (G) Medial-lateral pronation angle, which is the angle from the midpoint of the midfoot (the center of the triangle formed by the navicular marker (NAV) and the proximal metatarsal markers of the first and fifth metatarsals (PMT1 and PMT5)) to the midpoint of the forefoot ( The vertical upward direction of the plane formed by the parallel line of the connecting line of the first and fifth metatarsal distal marking points (DMT1, DMT5) is the rotation axis, and the angle between the projection of the inner foot axis formed by the proximal marking point of the first metatarsal (PMT1) to the distal marking point of the first metatarsal (DMT1) on the plane perpendicular to the rotation axis and the projection of the outer foot axis formed by the proximal marking point of the fifth metatarsal (PMT5) to the distal marking point of the fifth metatarsal (DMT5) on the plane perpendicular to the rotation axis;

[0089] (H) Ankle-to-calcaneal extension distance, specifically the distance between the lateral malleolus landmark (LCL) and the lateral calcaneal landmark (LCL).

[0090] The gait data feature relationship matrix involved in the feature space of this embodiment is obtained in the following way:

[0091] A feature relationship matrix between the gait data features in 8 dimensions and the 12 bony landmarks mentioned in S101 is established, and the matrix is ​​multiplied with its transposed matrix to obtain a relationship matrix of the gait data features in 8 dimensions, in which all diagonal elements are set to 0.

[0092] S103, establish a foot sports injury auxiliary diagnosis model and algorithm based on spatiotemporal Transformer network;

[0093] It should be noted that, in this embodiment, Figure 2 As shown, this method uses a spatiotemporal Transformer network model for auxiliary diagnosis of foot sports injuries. The network model has a chain structure. Each spatiotemporal Transformer network (ST Transformer Block) in the chain structure is repeated. The state of the unit is transmitted sequentially in the chain structure, and only a small amount of linear interaction is performed with the network. Information is easy to flow between ST Transformer block units. Therefore, the network structure has good temporal memory.

[0094] like Figure 3 As shown in the figure, each ST Transformer Block unit mainly consists of two parts, namely the spatial Transformer and the temporal Transformer modules.

[0095] like Figure 4As shown in the figure, the spatial Transformer module consists of a spatial-temporal position embedding layer, a fixed graph convolution layer, a dynamic graph convolution layer, and a gate mechanism for information fusion. Each spatial Transformer unit consists of 5 steps, including:

[0096] The output of the previous unit Input to the spatial position embedding unit (Positional Embedding) in the spatial Transformer module, through the spatial position matrix (Initialized according to the gait data relationship feature matrix) Learn the embedded features ;

[0097] Embedded Features The fixed graph convolution layer (GraphConvolution) based on Chebyshev polynomial approximation is sent to learn the structure-aware node features according to the learned weights and predefined graphs to obtain the structure-aware node features ;

[0098] The embedded feature matrix Through the corresponding weight matrix Projected into three potential subspaces (query subspace Q, key subspace K, value subspace V), the corresponding subspace feature vectors are obtained , and then the spatial dependency relationship of the nodes in the graph is obtained through the dot product, and finally the spatial dependency relationship is obtained through the spatial Sum value subspace vector Get node features ;

[0099] The node features to be updated Send it to the feed forward neural network structure to improve the learned node features For the conditional classification prediction, the improved features are obtained ;

[0100] The improved features and structure-aware node features The output of the spatial Transformer module is obtained by weighted fusion through the sigmoid gate structure. , and send it to the subsequent time Transformer module, that is, .

[0101] like Figure 5As shown in the figure, the temporal Transformer module consists of a spatial-temporal position embedding layer, a dynamic graph convolution layer, and a gate mechanism for information fusion. Each spatial Transformer unit consists of three steps, including:

[0102] The output of the spatial Transformer The temporal position embedding unit (Positional Embedding) input to the temporal Transformer module is obtained through the temporal position matrix (Initialized from one-hot time encoding) Learn embedded features ;

[0103] The embedded feature matrix Through the corresponding weight matrix Projected into three potential subspaces (query subspace Q, key subspace K, value subspace V), the corresponding subspace feature vectors are obtained , and then the time dependency of the nodes in the graph is obtained through the dot product, and finally the spatial dependency Sum value subspace vector Update node features ;

[0104] The node features to be updated Send it to the feed forward neural network structure to improve the learned node features For the conditional classification prediction, the improved features are obtained Output is sent to the next ST Transformer Block unit or classification layer, that is, ;

[0105] The auxiliary diagnosis algorithm of foot sports injuries based on spatiotemporal Transformer network is as follows:

[0106] Input: Sample set ; Sample set data dimension X_dim; STTN network parameters: adjacency matrix A; number of output channels in_channels; number of Transformer channels embed_size; number of training rounds epochs; number of Spatial-temporal block stacking layers num_layers; input time dimension T_dim; number of Transformer heads head.

[0107] Output: Classification result (foot ligament damage / no damage).

[0108] Step 1: Input the sample set and initialize the STTN network storage state;

[0109] Step 2: Aggregate the sample set input through a 1×1 convolutional layer;

[0110] Step 3: Integrate the input feature vector through the spatial information embedding layer to obtain the spatial embedding feature;

[0111] Step 4: Send the spatial embedding features to the fixed graph convolution layer and dynamic graph convolution layer of the spatial Transformer module to learn the corresponding spatial features;

[0112] Step 5: The spatial features learned in the dynamic graph convolution layer are used to update the parameters of the spatial Transformer module in the STTN network through a three-layer feedforward neural network;

[0113] Step 6: The spatial features learned by the two convolutional layers are fused through linear projection and sigmiod activation function;

[0114] Step 7: Integrate the fused spatial feature vector through the temporal information embedding layer to obtain the temporal embedding feature;

[0115] Step 8: Send the temporal embedding features into the dynamic graph convolution layer of the temporal Transformer module to learn the corresponding temporal features;

[0116] Step 9: The obtained time features are passed through a three-layer feedforward neural network to update the parameters of the temporal Transformer module in the STTN network;

[0117] Step 10: Integrate the spatial feature vector fused in the spatial Transformer module with the updated temporal feature in the temporal Transformer module to obtain the input spatiotemporal feature of the next spatiotemporal convolution module;

[0118] Step 11: Iterate based on the number of spatiotemporal convolution modules set in the network. If the next module is considered to be a spatiotemporal convolution module, enter step 3 to continue iterating; if the next module is a classification layer module, enter the classification layer module for classification;

[0119] Step 12: The classification layer uses two 1×1 convolutional layers to implement classification based on the spatiotemporal features extracted by the last spatiotemporal convolution module and outputs the classification results.

[0120] S104, constructing a validity verification method based on receiver operating characteristic curve;

[0121] It should be noted that, in this embodiment, the above S104 is specifically:

[0122] First, establish the confusion matrix of the decision model. The confusion matrix summarizes the records in the data set according to the real categories and the categories predicted by the decision model in the form of a matrix, where the rows of the matrix represent the real values ​​and the columns of the matrix represent the predicted values. The confusion matrix of this decision model is as follows:

[0123]

[0124] Among them, TP (True Positive): the number of normal foot ligaments predicted as normal; FN (False Negative): the number of normal foot ligaments predicted as injured; FP (False Positive): the number of foot sports injuries predicted as normal; TN (True Negative): the number of foot sports injuries predicted as injured.

[0125] Secondly, the receiver operating characteristic curve is constructed. Four data can be further calculated from the confusion matrix, including:

[0126] ① Accuracy (ACC), which is used to calculate the proportion of all correct judgments of the decision model to the total observations, where:

[0127]

[0128] ② Precision (PPV), used to calculate the proportion of correct predictions among all the results that the decision model judges to be normal for the foot ligaments, where:

[0129]

[0130] ③ Sensitivity (TPR) is used to calculate the proportion of correct predictions among all the results whose true values ​​are normal foot ligaments, among which,

[0131]

[0132] ④ Specificity (TNR), which is used to calculate the true value of the correct prediction ratio among all the results of foot sports injuries, among which,

[0133]

[0134] By combining the true positive rate and false positive rate at different critical points, and taking the sensitivity (TPR) of different cutoff points as the vertical axis, the specificity (FPR) ( ) as the horizontal axis, and the receiver operating characteristic curve was constructed. This curve reflects the relationship between sensitivity and specificity under different critical value conditions through the composition method.

[0135] Finally, the accuracy of the decision is determined by calculating the area under the ROC curve (AUC). The larger the AUC value, the higher the accuracy of the decision.

[0136] In summary, this embodiment aims to address the problem of shortage of intelligent and accurate diagnostic methods in the existing diagnosis and treatment of foot sports injuries, and provides an auxiliary diagnosis method and system for foot sports injuries based on gait data features. The foot and ankle motion information involved in the present invention comes from an optical motion capture system based on an infrared camera. On this basis, after data preprocessing, a gait data feature space is constructed for the injury group and the control group. Using the gait data features as a training set, an auxiliary diagnosis method and system for foot sports injuries and a method for verifying their effectiveness are developed, which effectively improves the level of accurate description of foot and ankle gait data features, realizes quantitative and accurate foot sports injury decisions, and provides a high-quality and efficient intelligent auxiliary tool for the diagnosis of foot sports injuries.

[0137] Third embodiment

[0138] See also Figure 5 This embodiment provides a foot sports injury auxiliary diagnosis system based on gait data features, and the foot sports injury auxiliary diagnosis system based on gait data features includes:

[0139] A gait motion information preprocessing module, which is used to filter the raw data of the collected gait motion information, automatically segment the gait cycle according to the spatial coordinate positions of the feature marker points based on the common characteristics of human gait and the actual motion conditions of the individual subjects, and perform gait cycle standardization processing on the motion information after filtering;

[0140] A gait data feature space construction module, wherein the gait data feature space construction module is used to calculate the gait data features of the injury group and the control group in 8 dimensions according to the gait cycle segmentation and standardization results, and to construct a gait data feature space and a gait data feature relationship matrix;

[0141] An injury auxiliary diagnosis module, which is used to construct a spatiotemporal Transformer network, and train the network using the gait data features of the injury group and the control group as a training set to achieve an auxiliary diagnosis function for foot sports injuries;

[0142] The method and system validity verification module are used to construct a confusion matrix, calculate the sensitivity and specificity of the foot sports injury auxiliary diagnosis model, form a receiver operating characteristic curve, and verify the validity of the decision method by the area under the receiver operating characteristic curve to verify the validity of the decision model.

[0143] The foot sports injury auxiliary diagnosis system based on gait data features of the present embodiment corresponds to the foot sports injury auxiliary diagnosis method based on gait data features of the above-mentioned embodiment, wherein the functions implemented by each module unit in the foot sports injury auxiliary diagnosis system based on gait data features of the present embodiment correspond one-to-one to each process step in the above-mentioned method; therefore, they will not be repeated here.

[0144] In addition, it should be noted that those skilled in the art should understand that the embodiments of the present invention can be provided as methods, devices, or computer program products. Therefore, the embodiments of the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0145] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0146] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable terminal device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0147] It should also be noted that, in this article, the terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0148] Finally, it should be noted that the above is a preferred embodiment of the present invention. It should be pointed out that although the preferred embodiment of the present invention has been described, once the basic creative concept of the present invention is known, a number of improvements and modifications can be made by ordinary technicians in this field without departing from the principles of the present invention. These improvements and modifications should also be regarded as the protection scope of the present invention. Therefore, the attached claims are intended to be interpreted as including the preferred embodiment, as well as all changes and modifications that fall within the scope of the embodiments of the present invention.

Claims

1. A foot sports injury auxiliary diagnosis system based on gait data features, characterized in that: The foot sports injury auxiliary diagnosis system based on gait data features includes: A data preprocessing module, which is used to filter the raw data of the collected gait motion information to reduce the noise of the raw motion information; based on the common characteristics of human gait and the actual motion conditions of the individual subjects, the filtered motion information is automatically segmented into gait cycles according to the spatial coordinate positions of the feature marker points, and an interpolation model is constructed to standardize the number of sampling points in the gait cycle; The collected gait motion information includes: Based on the Heidelberg foot measurement method, a foot and ankle motion measurement model covering the knee joint, ankle joint and foot is established; based on the foot and ankle motion measurement model, the motion information of the subject is collected through an optical motion capture system; wherein, The characteristic points of the knee joint part measurement include: the inner and outer marking points of the knee joint along the direction of the maximum distance of the knee joint bending axis; The characteristic points measured in the foot part include: the medial malleolus mark point along the direction of the maximum distance of the ankle bone rotation axis and the lateral malleolus mark point along the direction of the maximum distance of the ankle bone rotation axis; The characteristic points of foot measurement include: the most convex position marking points of the medial, lateral and dorsal sides of the calcaneus, the navicular marking point whose axis is about 45° to the ground when viewed from the sagittal plane, the proximal and distal marking points of the 1st and 5th metatarsal bones; A gait data feature space construction module, wherein the gait data feature space construction module is used to calculate the gait data features of the injury group and the control group in 8 dimensions according to the gait cycle segmentation and standardization results, construct a gait data feature space, and obtain a gait data feature relationship matrix; An injury auxiliary diagnosis module is used to construct a spatiotemporal Transformer network, and train the network using the gait data features of the injury group and the control group as a training set to achieve an auxiliary diagnosis function for foot sports injuries; A method and system validity verification module, wherein the method and system validity verification module is used to construct a confusion matrix, calculate the sensitivity and specificity of the foot sports injury auxiliary diagnosis model, form a receiver operating characteristic curve, and verify the validity of the decision method and the validity of the decision model by the area under the receiver operating characteristic curve; The constructing of the gait data feature space includes: The gait data features of foot sports injuries are extracted, and the extracted gait data features include: tibia-talar flexion angle, forefoot-heel abduction angle, medial arch angle of the foot, lateral arch angle of the foot, rotation angle, forefoot-ankle external rotation angle, medial-lateral foot mid-rotation angle, ankle-heel extension distance; For the relationship matrix between the above 8 gait data features and the 12 dorsal marker points of the root bones, the relationship matrix of 8 gait data features is extracted; The function of realizing auxiliary diagnosis of foot sports injuries includes: The gait data features of the injury group and the control group were used as the training set for the auxiliary diagnosis model of foot sports injuries; Construct a foot sports injury auxiliary diagnosis method based on the spatiotemporal Transformer network, input the training set into the network, and learn the mapping relationship from gait data features to injury decisions; Optimize the decision network parameters and iterate repeatedly until the output is a discrimination result that meets the loss function requirements and accuracy requirements; The foot sports injury auxiliary diagnosis system based on gait data features adopts a spatiotemporal Transformer network model for auxiliary diagnosis of foot sports injuries. The spatiotemporal Transformer network model has a chain structure, and each spatiotemporal Transformer network in the chain structure, namely, the ST Transformer Block structure is repeated; each S-T Transformer Block unit consists of two parts, namely, a spatial Transformer module and a temporal Transformer module; The spatial Transformer module is composed of a spatial-temporal position embedding layer, a fixed graph convolution layer, a dynamic graph convolution layer, and a gate mechanism for information fusion; each spatial Transformer unit consists of 5 steps, including: The output of the previous unit The spatial position embedding unit input to the spatial Transformer module is obtained through the spatial position matrix Learning embedded features ; Embedded Features The fixed graph convolution layer based on Chebyshev polynomial approximation is fed into the network to learn the structure-aware node features according to the learned weights and predefined graphs, and obtain the structure-aware node features. ; The embedded feature matrix Through the corresponding weight matrix Projected into three potential subspaces, including query subspace Q, key subspace K and value subspace V, the corresponding subspace feature vectors are obtained , and then the spatial dependency of the nodes in the graph is obtained through the dot product, and finally through the spatial dependency Sum value subspace vector Get node features ; The node features to be updated Send it into the feedforward neural network structure to improve the learned node features For the conditional classification prediction, the improved features are obtained ; The improved features and structure-aware node features The output of the spatial Transformer module is obtained by weighted fusion through the sigmoid gate structure. , and send it to the subsequent time Transformer module, that is, ; The temporal Transformer module is composed of a spatial-temporal position embedding layer, a dynamic graph convolution layer, and a gate mechanism for information fusion; each temporal Transformer unit consists of three steps, including: The output of the spatial Transformer The temporal position embedding unit input to the temporal Transformer module is obtained through the temporal position matrix Learn to get the embedded feature matrix ; Embedded Feature Matrix Through the corresponding weight matrix Projected into three potential subspaces, including query subspace Q, key subspace K and value subspace V, the corresponding subspace feature vectors are obtained , and then the time dependency of the nodes in the graph is obtained through the dot product, and finally the spatial dependency Sum value subspace vector Update node features ; Updated node features Send it into the feedforward neural network structure to improve the learned node features For the conditional classification prediction, the improved features are obtained Output is sent to the next ST TransformerBlock unit or classification layer, that is, .

Images