Diagnosis method and system for glenoid cavity and humeral head defect area

Abstract

The invention discloses a diagnosis method and system for glenoid cavity and humeral head defect areas, and belongs to the technical field of image processing, and the method comprises the steps: obtaining the medical image data of a glenoid cavity and a humeral head to be processed; inputting the medical image data into a preset segmentation model to obtain a segmentation result of the glenoid cavity and the humerus head; based on the segmentation result, calculating the defect area and defect proportion of the glenoid cavity; and determining the defect degree of the humerus head based on the defect proportion. According to the diagnosis method and system for the glenoid cavity and humeral head defect area, an efficient and accurate intelligent diagnosis auxiliary tool is provided for clinic, and the good application prospect and popularization value are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a diagnosis method and system for glenoid and humeral head defect area. BACKGROUND

[0002] Shoulder dislocation is a common traumatic disease in clinical practice, which often leads to bone damage of the humeral head and glenoid. Humeral head defect and glenoid defect, as two common pathologies of shoulder joint diseases, seriously affect the stability and function of the shoulder joint of patients, and bring great trouble to the daily life and motor ability of patients. With the increase of exercise intensity and the occurrence of traumatic accidents, the incidence of humeral head and glenoid defects is increasing year by year. Humeral head defect mainly manifests as local bone damage or loss, while glenoid defect manifests as partial or complete loss of glenoid. These defects can cause a series of subsequent problems such as joint degeneration and arthritis. Therefore, early detection and accurate assessment of the area of these defects can provide an important basis for clinical treatment and intervention.

[0003] Traditional diagnosis methods of humeral head and glenoid defects rely on X-ray, CT scan and MRI images, and the range and degree of damage are determined by manual analysis of professional doctors. However, these methods are easily disturbed by doctor's experience, image noise and complex morphology, resulting in low accuracy and efficiency of diagnosis, especially when dealing with a large number of images, the judgment of doctors may deviate, affecting treatment decisions. SUMMARY

[0004] The present application provides a diagnosis method and system for glenoid and humeral head defect area to solve the technical problem that traditional diagnosis methods of humeral head and glenoid defects rely on manual analysis to determine the range and degree of damage, thereby resulting in low accuracy and efficiency of diagnosis and affecting treatment decisions.

[0005] To solve the above technical problems, the present application provides the following technical solutions:

[0006] In one aspect, the present application provides a diagnosis method for glenoid and humeral head defect area, which comprises:

[0007] Obtaining medical image data of the glenoid and humeral head to be processed;

[0008] Inputting the medical image data into a preset segmentation model to obtain a segmentation result of the glenoid and humeral head;

[0009] Based on the segmentation result, calculating the defect area and defect proportion of the glenoid;

[0010] Based on the defect proportion, determining the defect degree of the humeral head.

[0011] Further, the segmentation model comprises a plurality of feature enhancement modules constituting a Transformer structure; from the second feature enhancement module, the output of each feature enhancement module is sequentially processed by an adaptive medical convolution module and an edge attention mechanism module respectively, and the obtained feature data is input into a segmentation head; the segmentation head realizes segmentation of the glenoid and humeral head based on the input multiple sets of feature data, and outputs the segmentation result of the glenoid and humeral head.

[0012] Further, the feature enhancement module comprises an overlapping block embedding module and a plurality of feature enhancement units; wherein the plurality of feature enhancement units are connected in series, and adjacent two feature enhancement units are connected through a down-sampling layer;

[0013] The overlapping block embedding module introduces overlapping patch encoding, and the input image is transformed by a convolution kernel size of 7x7 convolution layer, and each vector in the result represents a context-aware overlapping region embedding as a Token sequence input into the Transformer structure; the input sequence x input into the feature enhancement unit is first processed by a feedforward feedback network to obtain an output x', which is expressed by the formula:

[0014] x' = x + γ Conv 1×1 (LN(x)) + FFN(x)

[0015] Wherein, γ is a learnable scaling parameter; Conv 1×1 is a local convolution feedback channel for embedding spatial structure information; LN(x) represents normalizing the input; FFN represents transforming each Token independently, and the transformation operation is expressed as:

[0016] FFN(x) = W2 σ(W1x + b1) + b2

[0017] Wherein, W1, W2 represent weight matrices, σ represents a nonlinear activation function for normalizing the input; b1 and b2 represent bias vectors;

[0018] x' then enters a dimension interaction module to enhance the direct information exchange between Tokens in different positions; then the x' processed by the dimension interaction module is divided into G groups in the channel dimension through a cascaded group attention mechanism module, and then local self-attention is independently executed for each group, and the computational complexity is reduced;

[0019] Finally, the outputs of each group are concatenated and subjected to information fusion operation as the output of the feature enhancement unit.

[0020] Further, the number of feature enhancement units is 3; in the first feature enhancement unit, G is set to 16, in the second feature enhancement unit, G is set to 8, and in the last feature enhancement unit, G is set to 4.

[0021] Further, the operation formula of the local self-attention is represented as:

[0022]

[0023] wherein Q (i) represents the i-th Query matrix; K (i) is the i-th Key matrix; V (i) is the i-th Value matrix; d k is the dimension of the Key; T represents the transpose of the matrix; Z (i) represents the output result of the i-th attention head; and Softmax() represents the normalization processing.

[0024] Further, the data processing process of the edge attention mechanism module comprises:

[0025] The first-order gradient information of the input image is extracted in the horizontal direction and the vertical direction respectively by using the Sobel operator; wherein the horizontal direction operator G x and the vertical direction operator G y are represented as:

[0026]

[0027] The convolution operation is performed on each channel of the input image and G x respectively to obtain the horizontal direction gradient map, and the convolution operation is performed on each channel of the input image and G y respectively to obtain the vertical direction gradient map;

[0028] The horizontal direction gradient map and the vertical direction gradient map are processed by the global average pooling layer respectively to obtain the aggregation tensor T H in the vertical direction and the aggregation tensor T W in the horizontal direction of the input image;

[0029] T H and T W are multiplied to obtain a position-related attention map M;

[0030] M is input into the 1x1 convolution layer, and then the output of the 1x1 convolution layer is added to the input image of the edge attention mechanism module to obtain the optimized feature map.

[0031] Further, the data processing process of the segmentation head comprises:

[0032] The input multiple sets of feature data are fused to obtain the fused feature data F fused ;

[0033] The weight of each pixel is dynamically calculated, wherein the weight generation formula is:

[0034] W = sigma(Conv(F fused ))

[0035] wherein W is a dynamically generated weight mask; sigma represents a nonlinear activation function for normalizing input; Conv(·) represents a convolution operation; the range of W is [0, 1], and the greater the value, the more the model focuses on the corresponding area;

[0036] According to W, a final dynamic mask is generated for guiding segmentation, wherein the dynamic mask generation formula is:

[0037] M dynamic = W o P

[0038] wherein M dynamic is a dynamic mask; P is a preliminary prediction result of the model; o represents pixel-by-pixel weighting;

[0039] Based on the generated dynamic mask, the glenoid and humeral head are segmented.

[0040] Further, based on the segmentation result, the defect area and defect ratio of the glenoid are calculated, including:

[0041] Using the segmentation model, a standard complete morphology of the glenoid is generated according to training data of healthy samples;

[0042] Using the segmentation model, an actual morphology of the glenoid is generated according to the medical image data;

[0043] The pixel area difference between the standard complete morphology and the actual morphology is calculated to obtain the defect area of the glenoid;

[0044] The percentage of the defect area in the complete glenoid area is calculated to obtain the defect ratio of the glenoid.

[0045] Further, based on the defect ratio, the defect degree of the humeral head is determined, including:

[0046] If the defect ratio is less than 10%, it is determined as slight defect;

[0047] If the defect ratio is between 10% and 30%, it is determined as moderate defect;

[0048] If the defect ratio is greater than 30%, it is determined as severe defect.

[0049] On the other hand, the present application also provides a diagnostic system for glenoid and humeral head defect area, which comprises:

[0050] a data acquisition module configured to acquire medical image data of a glenoid and a humeral head to be processed;

[0051] a target segmentation module configured to input the medical image data into a preset segmentation model to obtain a segmentation result of the glenoid and the humeral head;

[0052] a data calculation module configured to calculate a defect area and a defect ratio of the glenoid based on the segmentation result;

[0053] a defect degree judgment module configured to determine a defect degree of the humeral head based on the defect ratio.

[0054] In still another aspect, the present application also provides an electronic device comprising a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above method.

[0055] In still another aspect, the present application also provides a computer-readable storage medium, which stores at least one instruction, which is loaded and executed by the processor to implement the above method.

[0056] The technical scheme provided by the present application has at least the following beneficial effects:

[0057] The present application introduces an edge attention mechanism module and a feature fusion segmentation head for the complex morphology of defects, thereby optimizing the segmentation accuracy of the humeral head region. The feature enhancement module is used to improve the feature extraction efficiency and solve the problem of complex morphology. Through the cooperative task driving mechanism, the joint optimization of the double damage recognition task is realized, and the overall accuracy and robustness of the model in target recognition and area calculation are improved. Experimental results show that the network proposed in the present application performs well in the automatic detection and quantitative analysis of humeral head and glenoid defects, especially in complex background interference and small area lesion recognition, and exhibits good generalization ability. It provides an efficient, accurate and intelligent diagnostic aid tool for clinical application, and has good application prospect and popularization value. BRIEF DESCRIPTION OF DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0059] Figure 1 is the diagnostic method flowchart for the glenoid and humeral head defect area provided by the embodiment of the present application;

[0060] Figure 2Schematic diagram of the DDDNet model structure provided by an embodiment of the present invention;

[0061] Figure 3 Schematic diagram of the structure of the EGAM module provided by an embodiment of the present invention;

[0062] Figure 4 Schematic diagram of the OptiViT network module structure provided by an embodiment of the present invention;

[0063] Figure 5 Schematic diagram of the structure of the ASFFHead segmentation head provided by an embodiment of the present invention;

[0064] Figure 6 This is a schematic diagram of a label case provided by an embodiment of the present invention;

[0065] Figure 7 The following are the glenoid segmentation renderings provided by an embodiment of the present invention; the first column is the glenoid MRI image, the second column is the expert-labeled image, and the remaining columns are the segmentation renderings of various segmentation networks.

[0066] Figure 8 The following are the segmentation renderings of the humeral head provided by an embodiment of the present invention; wherein the first column is the humeral head MRI image, the second column is the expert-labeled image, and the remaining columns are the segmentation renderings of various segmentation networks;

[0067] Figure 9 is a structural diagram of an adaptive medical convolution module provided by an embodiment of the present invention;

[0068] Figure 10 This is a system block diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0069] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0070] First, it should be noted that in the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "exemplarily" is intended to present concepts in a concrete manner. In addition, in the embodiments of the present invention, the meaning of "and / or" can be both or either of the two.

[0071] First embodiment

[0072] The embodiment provides a diagnostic method for glenoid and humeral head defect area, which can be implemented by an electronic device, which can be a terminal or a server.

[0073] The execution flow of the method is as shown in Figure 1 The method comprises the following steps:

[0074] S1, obtaining medical image data of a glenoid and a humeral head to be processed;

[0075] S2, inputting the medical image data into a preset segmentation model to obtain a segmentation result of the glenoid and the humeral head;

[0076] Specifically, the structure of the segmentation model DDDNet of the embodiment is as shown in Figure 2 It is composed of an Optimized Vision Transformer-based Feature Enhancement Module (OptiViT) of a Transformer structure, which enhances the high-order features of each stage through an adaptive medical convolution module, uses an Edge-global Attention Module (EGAM) to utilize the detailed features of the shallow layer to improve the edge segmentation effect, and further improves the performance of the DDDNet through a feature fusion segmentation head (Adaptively Spatial Feature Fusion Detection Head, ASFFHead) to fuse multi-scale feature maps.

[0077] Compared with the traditional segmentation model, the DDDNet effectively maintains high-precision segmentation when processing the target and complex boundary in the glenoid and humeral head image. The DDDNet can meet the real-time demand in medical diagnosis through its lightweight design and efficient inference process, helping doctors to obtain accurate diagnostic results in a short time. The DDDNet has precise segmentation capability and efficient real-time processing capability, and the DDDNet can improve the diagnostic accuracy of orthopedic diseases and has excellent precision and efficiency in the glenoid and humeral head defect segmentation task.

[0078] Next, the main functional modules in the DDDNet are described in detail.

[0079] 1. Edge attention mechanism

[0080] EGAM is an innovative module combining edge information attention mechanism, and the specific structure is as shown in Figure 3As shown, the purpose is to improve the attention ability of the edge region of the deep learning model in the visual task. The subtle defects and edge key regions of the glenoid and humeral head have very important diagnostic value in medical image analysis and target segmentation tasks. By combining and fine-tuning the global attention mechanism and the edge attention mechanism, EGAM can effectively enhance the sensitivity of the model to the edge region of the humeral head defect, and improve the detection and segmentation accuracy.

[0081] EGAM enhances the sensitivity of the deep learning model to the key region by combining edge information and global features, as shown in Figure 3 As shown, C, H, and W represent the width, height, and channel number of the input image, respectively. The edge region usually contains important structural information in the image, and the outline and shape of the defect in the medical image are often located in the edge region. EGAM extracts edge features to enable the model to focus on these detailed regions. First, the Sobel operator is used to extract the first-order gradient information of the image in the horizontal and vertical directions, respectively. The horizontal direction operator G x and the vertical direction operator G y are expressed as:

[0082]

[0083] For a given input image Each channel is respectively convolved with G x , G y to obtain the gradient map in the horizontal and vertical directions, and the two are spliced to generate the edge map The edge map highlights the edge region in the image. The edge information is encoded into a significant feature map through a convolutional neural network layer, providing edge context information for subsequent steps.

[0084] EGAM uses global context information to capture important regions in the image and calculates channel-level attention weights based on global features. Spatial average pooling is performed on each channel of the input feature map F to obtain the global context information of the channel. The average pooling operation is as follows:

[0085]

[0086] Where F c,i,j represents the pixel value of the cth channel in the ith row and jth column of the input feature map.

[0087] Then, the module calculates the aggregation tensors T H and T WThese two tensors converge on the pixel-level information of the feature map in space, forming a weight distribution based on spatial position, which represents the importance of different positions and helps to highlight edge regions and detailed information, thereby enhancing the model's sensitivity to edges. Then, T H and T W are multiplied to obtain a position-dependent attention map M, which can further guide the network to focus on the edge regions in the image. Finally, M is passed through a 1x1 convolution layer and added to the input feature map to obtain the optimized feature map. This process effectively improves the segmentation of high-complexity and low-contrast edge regions in images by enhancing the feature response of edge regions, better reconstructing details, and improving the accuracy and robustness of segmentation.

[0088] 2. OptiViT module

[0089] Medical images often have high resolution and more redundant information in segmentation tasks. Directly using standard convolution not only has large computational overhead, but also may cause excessive smoothing of features. To speed up the extraction of damage features, the OptiViT network is designed as the network, and to better extract local information in the glenoid damage image, the specific network structure is as shown in Figure 4 .

[0090] The morphology of the defect is complex, and its anatomical features are characterized by the coexistence of continuity in local regions and irregularity in defect regions. The overlapping patch encoding is introduced, and the input image is transformed by a 7x7 convolution layer with a convolution kernel size, and each vector in the result represents a context-aware overlapping region embedding as a Token sequence input to the Transformer. The input sequence x input to the OptiVit unit is first passed through a feedforward feedback network, and the specific operation is formula (1) to obtain the output x'.

[0091] x' = x + γ · Conv 1×1 (LN(x)) + FFN(x) (1)

[0092] where γ is a learnable scaling parameter, Conv 1×1 is a local convolution feedback channel used to embed spatial structure information, LN(x) represents normalization of the input, and FFN represents independent transformation of each Token, and the transformation operation is formula (2).

[0093] FFN(x) = W2 · σ(W1x + b1) + b2 (2)

[0094] where W1, W2 represent weight matrices, and σ represents a nonlinear activation function for input normalization; b1 and b2 represent bias vectors, which are learnable translation parameters. Subsequently, Token Interantion (which is referenced from the Token-to-Token ViT model, which realizes explicit interaction modeling between tokens through multi-step token conversion, effectively improving global modeling capability) is entered to enhance direct information exchange between tokens in different positions. A cascaded group attention mechanism module is embedded in the Transfomer structure, which divides the input feature image into G groups in the channel dimension to obtain feature Figure X .

[0095]

[0096] In the first OptiVit unit, G is set to 16, in the second OptiVit unit, G is set to 8, and in the last OptiVit unit, G is set to 4.

[0097] Local self-attention is independently performed for each group, and the computational load is reduced. The local attention operation formula is as follows:

[0098]

[0099] where Q (i) represents the i-th Query matrix, K (i) is the i-th Key matrix, V (i) is the i-th Value matrix, d k is the dimension of the Key; T represents the transpose of the matrix; Z (i) represents the output result matrix of the i-th attention head, which fuses the information of the current Query and all Key-Value pairs; Softmax( ) represents the normalization processing of attention scores, which is converted into a probability distribution to highlight the weight of key matching items, in order to highlight important token pairs. After concatenating the outputs of each group, information fusion operation is performed, and the output is Z = Concat(Z (1) ,...,Z (G)); then fuse other group information to enhance the global expression ability. After adding OptiViT Block in the backbone network, the model can extract the glenoid features from both local and global perspectives. On the one hand, by capturing small-scale detail information, it can accurately separate the boundary between the complete area and the defect area. On the other hand, through the parallel processing of multiple groups of convolutions, the model can pay attention to the overall morphological features of the glenoid, which can effectively distinguish the significant differences between normal and abnormal morphologies. OptiViT strengthens the high-resolution representation ability of the glenoid edge and circular shape and the defect area in the feature extraction stage of the segmentation model, and can still output stable segmentation results in the case of high image noise or incomplete glenoid morphology.

[0100] 3. ASFFHead segmentation head

[0101] ASFFHead plays an important role in the defect area detection task. The humeral head and glenoid may be damaged to varying degrees, and accurate defect area calculation and accurate defect segmentation depend on the effective fusion of multi-scale information. As shown in Figure 5 , ASFFHead optimizes the model's response to different scale features to improve the detection accuracy of humeral head defects by its feature aggregation and selective fusion mechanism when processing humeral head defects of different sizes and morphologies.

[0102] Noise and complex background are common problems in medical images, which makes the boundary of humeral head and glenoid defects blurred. The selective fusion mechanism of ASFFHead allows the network to dynamically select the most relevant features for fusion according to the importance of the defect area. In this way, the model can focus on the defect part of the humeral head and glenoid in complex background and balance between global and local features to improve the accuracy of defect detection. Since the morphology of the humeral head is usually approximately circular, ASFFHead can help the model effectively capture the regularity of this morphology and accurately segment and identify the bone defects and fractures in the image.

[0103] Shallow features can reflect boundary contours and texture information, while deep features focus on high-level semantic expression. Both are integrated through the feature fusion module to achieve comprehensive representation of the defect area. The feature fusion formula is:

[0104] F fused = Concat(F shallow ,F deep ) (5)

[0105] where F shallow and F deeprespectively, and Concat(*) represents the feature concatenation operation. The key of dynamic mask generation is to adjust the weight of the region of interest according to the specific characteristics of the input image. By constructing a weight generation network, the weight of each pixel is dynamically calculated, so as to highlight the significant areas of the damage area and glenoid shape. The weight generation formula is:

[0106] W = σ(Conv(F fused )) (6)

[0107] where W is the dynamically generated weight mask, σ represents a nonlinear activation function, which is used to normalize the input, and Conv(·) represents a convolution operation. The weight W ranges from 0 to 1, and the greater the value, the more the model focuses on the area. According to the dynamically generated weight W, the module generates the final dynamic mask for guiding segmentation. The dynamic mask generation formula is:

[0108] M dynamic = W ⊙ P (7)

[0109] where M dynamic is the dynamic mask, P is the preliminary prediction result of the model, and ⊙ represents a pixel-wise weighting operation. Multi-scale objects are usually processed in traditional target detection models through a pyramid structure or shallow-to-deep feature fusion. This method often has the problems of loss of detailed information and insufficient accuracy of response to different scale objects. ASFFHead effectively overcomes these challenges by enhancing the aggregation ability of multi-level features and adopting a selective fusion strategy. It improves the detection accuracy of large-scale objects by utilizing different levels of feature maps, and better captures the details of small-scale objects to improve the overall detection effect.

[0110] 4. Adaptive medical convolution module

[0111] This module integrates three branch structures of lightweight channel transformation, depth separable convolution and structure-aware convolution, and realizes precise modeling and feature enhancement of irregular targets such as shoulder cysts through multi-scale and multi-path information extraction mechanism. The specific structure is as shown in Figure 9 .

[0112] In this module, the input feature map will be sent into three parallel branches at the same time to extract complementary features, which are optimized and designed for channel dimension compression and enhancement, local structure modeling, and boundary sensitivity modeling, respectively.

[0113] First is the lightweight channel transformation branch, which uses standard convolution operations to expand the input feature map in the channel dimension, and further enhances the non-linear expression ability through batch normalization and activation function. This branch is mainly used to capture important information in the channel dimension, strengthen the network's modeling of high-value semantic features, and reduce the interference of redundant channels on model learning. The expression formula of the above process is as follows:

[0114] F1 = SiLU(BN(W1*X)) (8)

[0115] Where F1 represents the feature map obtained by lightweight channel transformation; W1 is a 1x1 convolution kernel, BN represents the normalization operation, SiLU is the activation function, and the size of the final output feature map is WxHxC.

[0116] The second branch uses depth separable convolution, and the padding and stride values are expanded to 1 for more detailed feature mining, so as to capture medium-scale cyst features. The expression after the channel-by-channel convolution of the depth separable convolution is as follows:

[0117]

[0118] Where D c (i,j) represents the value of the c-th channel convolution output at position (i,j); K c (m,n) represents the convolution kernel weight of the c-th channel, (m,n) is the index of the position in the convolution kernel; X c (i+m,j+n) represents the pixel value of the input feature Figure X c at position (i+m,j+n); i,j represents the spatial coordinate index in the current output feature map.

[0119] The weight K c of all positions in the 3x3 convolution window is multiplied by the input X c pixel by pixel and summed up to obtain each channel feature map D c , and then the channel feature maps are fused by pixel-by-pixel convolution. The operation formula is as follows:

[0120] F2 = SiLU(BN(W2*D)) (10)

[0121] Where W2 represents the weight matrix used for linear transformation; D represents the feature map set obtained by concatenating all channel convolution results.

[0122] The feature map F2 of the second branch is obtained, and the size of the feature map F2 is still WxHxC. The third branch introduces a Snake convolution with strong structure perception. The Snake convolution can adaptively adjust the sampling path based on the gradient information of the feature map, dynamically slide along the target edge to extract features, and has stronger adaptability and resolution when processing irregular edges, curved structures and special-shaped lesion areas. The formula of the above operation is shown in equation (11), and the normalization and activation function processing are performed after the convolution operation, as shown in equation (12).

[0123]

[0124] F3 = SiLU(BN(SnakeConv(X))) (12)

[0125] wherein F3(i,j) represents the output value of the Snake convolution at position (i,j); K(m,n) represents the weight parameter of the convolution kernel at position (m,n); Ω represents the index set of the convolution kernel window; X(i+Δx m,n ,j+Δy m,n ) represents the pixel value at the offset position i+Δx m,n ,j+Δy m,n of the input feature map; Δx m,n represents the horizontal offset corresponding to the position (m,n) in the convolution kernel; Δy m,n represents the vertical offset corresponding to the position (m,n) in the convolution kernel; i,j represents the spatial coordinate index in the current output feature map; F3 represents the complete feature map composed of the Snake convolution outputs at all positions; SnakeConv(X) represents the intermediate result obtained by performing the Snake convolution operation on the input feature Figure X ;

[0126] After the outputs of the three branches are spliced in the channel dimension, a feature map with a size of WxHx3C is obtained, and finally a 1x1x2C convolution is performed to complete information fusion and compression. The final output is a feature map with fused multi-dimensional information, and the size of the final output is WxHx2C.

[0127] S3, based on the segmentation result, calculating the glenoid defect area and defect ratio;

[0128] Specifically, the core idea of the defect area and defect ratio calculation is to predict the complete shape and actual shape of the glenoid through the deep learning model, calculate the pixel area difference between the two, and then obtain the absolute area and relative percentage of the glenoid defect. The normal glenoid shape is generated by the deep learning model according to the training data. The glenoid shape in the current state of the patient is segmented from the actual image by the deep learning model. The defect pixel area is the pixel area difference between the complete glenoid and the actual glenoid. The defect ratio is the percentage of the defect area to the complete glenoid area. In order to quantitatively evaluate the defect degree of the glenoid, the algorithm calculates the absolute area and relative percentage of the defect based on the complete glenoid shape and the actual glenoid shape predicted by the deep learning model. This calculation process combines deep learning segmentation results and mathematical geometric analysis, which can accurately quantify the damage degree and provide data support for subsequent clinical diagnosis and treatment.

[0129] Through the deep learning model, the standard complete shape of the glenoid is generated using the training data of healthy samples, denoted as M full , wherein:

[0130] M full (x,y)∈{0,1}

[0131] , wherein M full (x,y)=1 indicates that the pixel (x,y) belongs to the complete bone region.

[0132] The actual shape of the glenoid or humeral head extracted from the current image of the patient is denoted as M actual , wherein:

[0133] M actual (x,y)∈{0,1}

[0134] , wherein M actual (x,y)=1 indicates that the pixel (x,y) belongs to the complete glenoid region.

[0135] The defect area pixel area can be calculated by the pixel difference between the complete glenoid shape and the actual glenoid shape:

[0136]

[0137] The defect ratio is the percentage of the defect area to the complete glenoid area, denoted as:

[0138]

[0139] S4, based on the defect ratio, determining the defect degree of the humeral head;

[0140] Specifically, according to the calculated defect area, the embodiment defines three levels of defect severity: slight defect, moderate defect, and severe defect. In order to determine the threshold of these levels, we divide the humeral head defects according to experience and medical standards.

[0141] Slight defect (Grade 1):

[0142] The ratio of defect area to complete area is less than 10%, that is, P defect < 0.1, it is a slight defect.

[0143] Moderate defect (Grade 2):

[0144] The ratio of defect area to complete area is between 10% and 30%, that is, 0.1 ≤ P defect < 0.3, it is a moderate defect.

[0145] Severe defect (Grade 3):

[0146] The ratio of defect area to complete area is greater than 30%, that is, P defect ≥ 0.3, it is a severe defect.

[0147] Through the above calculation method, the doctor can accurately understand the injury degree of the patient, thereby providing a quantitative basis for the treatment plan. In addition, this method also has good universality and can be popularized and applied to other similar medical image segmentation and quantitative analysis tasks.

[0148] Next, the effectiveness and superiority of the method of the present application are verified through experiments.

[0149] 1. Introduction of data experiment set and experimental environment

[0150] In this embodiment, the experimental data set is provided by a cooperative hospital, which contains 800 MRI images of shoulder joint glenoid and 792 MRI images of shoulder humeral head. These images are collected by using an ultrasonic machine with appropriate probe frequency, ensuring the quality and applicability of the images. The labeling work of the data set is completed by two experienced sports medicine surgeons using the labelme image labeling software, and is strictly checked by a kinematic injury expert to ensure the accuracy of the labeling. The label case is shown as Figure 6 .

[0151] In order to effectively train and evaluate the deep learning model, the experimental data set is divided into training set, validation set and test set according to the ratio of 8:1:1. 640 glenoid MRI and 633 humeral head MRI images are used as training set for model training and optimization, and the remaining images are 1:1 for validation set and test set.

[0152] 2. Evaluation index

[0153] To evaluate the performance of the algorithm in the diagnosis of defect area, various standard segmentation evaluation indicators are adopted in this embodiment to show the performance of the network. In this work, four indicators are selected to evaluate the performance, including intersection over union (IoU), Dice similarity coefficient (DICE), precision (PRE), HD distance (Hausdorff, HD), and parameter quantity (Params). These indicators not only help measure the performance of the model in segmentation accuracy, but also provide in-depth analysis of segmentation boundaries, computational efficiency, and other aspects. These indicators are described as follows:

[0154]

[0155] HD = max{dis(P, G), dis(G, P)}

[0156] Where A represents the predicted region, B represents the true region, TP represents the true positive, FP represents the false positive, dis(P, G) represents the farthest distance from the predicted boundary point set P to the true boundary point set G, and dis(G, P) represents the farthest distance from the true boundary point set G to the predicted boundary point set P.

[0157] Where the intersection over union and the Dice similarity coefficient are the most commonly used evaluation indicators in image segmentation tasks, and are widely used in medical image segmentation tasks. IoU calculates the intersection and union ratio of the predicted region and the true label region, which can intuitively reflect the accuracy of the model in the target region overlap. The Dice coefficient is another evaluation indicator for region overlap, especially suitable for tasks that focus on small target regions, and is often used to evaluate the segmentation effect of organ or lesion regions in medical images. In this embodiment, these two indicators are used to evaluate the segmentation accuracy of the model in the humeral head defect region. In addition, considering that the humeral head defect image in this embodiment has a complex shape and boundary, the Hausdorff distance (HD) is used to evaluate the maximum distance between the predicted boundary and the true boundary. This indicator is particularly effective in handling small-scale, complex detail segmentation tasks, helping us understand the performance of the model in boundary accuracy. Precision, as a classic evaluation standard, is also used in this embodiment to measure the segmentation accuracy and coverage of the model. Precision reflects the proportion of true positive examples in the predicted positive region. These indicators can help us balance the false positive and false negative examples of the model, and have important significance in the segmentation task of the target region.

[0158] 3. Comparative experiment

[0159] In this embodiment, in order to verify the effect of the proposed feature enhancement module-based humeral head defect area diagnosis algorithm, we conducted comparative experiments with several current mainstream medical image segmentation models. These models include, (1) U-Net as a classic model in the field of medical image segmentation, it has shown good performance in many medical image tasks. It extracts and reconstructs the details of the target area through an encoder-decoder structure, which is suitable for relatively simple segmentation tasks. We compared U-Net with the model proposed in this study to evaluate its accuracy and detail capture ability in processing glenoid defect images. (2) Mask R-CNN is a region proposal network (RPN) based instance segmentation model with strong segmentation accuracy and boundary detection ability. Since humeral head defect images often contain complex boundaries and small targets, Mask R-CNN will be one of the comparison models to verify whether the proposed model is more advantageous in handling such tasks. (3) DeepLabV3+ uses techniques such as dilated convolutions (Dilated Convolutions) and spatial pyramid pooling (ASPP) to effectively capture context information at different scales. This model performs well in complex scenarios and is suitable for tasks that require multi-scale feature extraction. Therefore, it is an important benchmark model in our comparative experiments. (4) Yolov8 by comparing with these models, this embodiment will deeply analyze the performance differences of each model in the glenoid defect segmentation task, especially in terms of segmentation accuracy, boundary capture, computational efficiency, etc. (5) YOLOv11 is a newer version of the YOLO series that further improves target detection accuracy and inference speed. This version integrates the latest neural network architecture and training techniques to optimize target positioning and multi-class object detection capabilities. Since humeral head defect images often contain complex boundaries and smaller lesion areas, YOLOv11 will be one of the comparison models to verify the relative performance of the proposed model in handling these complex scenarios. The experimental results will provide strong support for the advantages of the model proposed in this study and verify its feasibility and practicality in clinical applications. Five evaluation indicators, including, coefficient, precision, HD distance, and running parameter quantity, are used to quantitatively analyze the models.

[0160] Table 1 Comparison results of different algorithms on the shoulder defect dataset

[0161]

[0162]

[0163] The experimental results in Table 1 show that the model proposed in this invention outperforms other advanced models in almost all evaluation indicators. In the segmentation of lesion images, this model achieved the best IoU (85.51%), accuracy (92.41%) and Hausdorff distance (1.077mm). This achievement is mainly due to the introduction of a hierarchical cascade feature fusion module in the upsampling part, which can effectively aggregate multi-scale information, thereby improving the accuracy of segmentation; at the same time, the improved dynamic mask module helps the model focus on key information, especially in the segmentation of sacrocarpal joints with complex backgrounds, showing significant advantages. In addition, although the segmentation effect has been significantly improved, the number of model parameters has been effectively controlled, even lower than other similar methods. This advantage is attributed to the introduction of partial convolution, which significantly reduces the number of floating-point operations and memory access, thereby improving computational efficiency.

[0164] 4. Ablation Experiment

[0165] After introducing the EG-Module, the model's IoU increased to 87.54%, while DICE decreased slightly to 93.77%. However, PRE increased to 96.34%, and HD decreased to 2.4 mm. This demonstrates that the module helps enhance the depiction of local details in defect regions and improves prediction accuracy. Further adding the ASFFHead increased the IoU to 88.42%, DICE to 93.14%, and PRE significantly increased to 97.29%. The HD decreased to 2.2 mm, the lowest value to date. This demonstrates that the module effectively integrates feature information at different scales, particularly excelling in edge segmentation. When all three modules are integrated to form DDDNet, the IoU reaches 89.77%, DICE increases to 94.71%, PRE is 98.16%, and HD remains at 2.3 mm. This overall performance outperforms all individual combinations, fully demonstrating the complementary and synergistic enhancement effects of each module and providing strong support for refined medical image segmentation.

[0166] Table 2 Ablation experiment results

[0167]

[0168]

[0169] 5. Qualitative results

[0170] Effect of glenoid division Figure 7 As shown in the figure, the first column is the glenoid MRI image, the second column is the expert-labeled image, and the rest are the segmentation effect diagrams of various segmentation networks. Compared with other algorithms, the segmentation results of the algorithm in this paper have finer contours and are closer to the expert-labeled images.

[0171] The effect of humeral head segmentation is as follows Figure 8As shown in the table, the first column is the humeral head MRI image, the second column is the expert-labeled image, and the remaining columns are the segmentation results of various segmentation networks.

[0172] Second embodiment

[0173] The present embodiment provides a diagnosis system for glenoid and humeral head defect area, which comprises the following modules:

[0174] A data acquisition module is configured to acquire medical image data of a glenoid and a humeral head to be processed.

[0175] A target segmentation module is configured to input the medical image data into a preset segmentation model to obtain a segmentation result of the glenoid and the humeral head.

[0176] A data calculation module is configured to calculate a defect area and a defect ratio of the glenoid based on the segmentation result.

[0177] A defect degree judgment module is configured to determine a defect degree of the humeral head based on the defect ratio.

[0178] It should be noted that the diagnosis system for glenoid and humeral head defect area of the present embodiment corresponds to the diagnosis method for glenoid and humeral head defect area of the first embodiment described above; wherein the functions realized by each functional module in the diagnosis system for glenoid and humeral head defect area of the present embodiment correspond one-to-one to each process step in the diagnosis method for glenoid and humeral head defect area of the first embodiment; therefore, no further description is given here.

[0179] Third embodiment

[0180] The present embodiment provides an electronic device, such as Figure 10 As shown in the table, the electronic device comprises a processor and a memory; wherein the processor and the memory can be connected through a communication bus; the memory stores at least one instruction, which is loaded and executed by the processor to implement the method of the first embodiment described above. In addition, the electronic device can further comprise a transceiver, and the processor and the transceiver can be connected through a communication bus, and the transceiver is configured to communicate with other devices.

[0181] Next, the various components of the electronic device will be described in detail: Figure 10

[0182] ​Among them, the processor is the control center of the electronic device, and the electronic device may include multiple processors, each of which may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor here can be a processor or a general term for multiple processing elements. For example, the processor is one or more central processing units (CPUs), or other general-purpose processors, application specific integrated circuits (ASICs), or one or more integrated circuits configured to implement an embodiment of the present invention, such as one or more microprocessors (digital signal processors, DSPs), or one or more field programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor can perform various functions of the electronic device by running or executing software programs stored in the memory and calling data stored in the memory.

[0183] In a specific implementation, as an embodiment, the processor may include one or more CPUs, such as Figure 10 The CPU0 and CPU1 shown in FIG are, of course, only exemplary.

[0184] The memory is used to store the software program for executing the solution of the present invention, and the execution is controlled by the processor. The specific implementation method can refer to the above method embodiment and will not be repeated here.

[0185] Optionally, the memory may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may be integrated with the processor or exist independently and accessed through the interface circuit ( Figure 10 (not shown) is coupled to the processor, which is not specifically limited in this embodiment of the present invention.

[0186] The transceiver may include a receiver and a transmitter ( Figure 10 The receiver is used to implement the receiving function, and the transmitter is used to implement the sending function. The transceiver can be integrated with the processor or exist independently and communicate with the electronic device through the interface circuit ( Figure 10 (not shown) is coupled to the processor, which is not specifically limited in this embodiment of the present invention.

[0187] In addition, it should be noted that Figure 10 The structure of the electronic device shown in the figure does not constitute a limitation on the device. The actual device may include more or fewer components than shown, or may combine certain components, or arrange the components differently. In addition, the technical effects achieved by the electronic device when executing the method of the first embodiment can refer to the technical effects described in the first embodiment above, and therefore will not be repeated here.

[0188] Fourth embodiment

[0189] This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device. The instructions stored therein can be loaded by a processor in a terminal to execute the method described above.

[0190] Moreover, it should be noted that the present application can be provided as a method, an apparatus, or a computer program product. Therefore, the embodiments of the present application can take the form of an entirely or partially hardware embodiment, an entirely or partially software embodiment, or an embodiment combining software and hardware aspects. Furthermore, when implemented in software, the embodiments of the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, a computer diskette, an optical storage medium, a magnetic storage medium, and a semiconductor memory device). The computer program product includes one or more computer instructions that when loaded and executed by a computer, cause the computer to carry out the processes or functions described in the embodiments of the present application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, such as from a website, a computer, a server, or a data center to another website, computer, server, or data center through a wired (for example, infrared, wireless, microwave, or the like) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, data center, or the like, including one or more collections of available media. The available media can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0191] The embodiments of the present application 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 application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams 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 produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate a device that implements the flow Figure 1 The flow or the plurality of flows and / or blocks Figure 1 The device that implements the functions specified in the flow or the plurality of flows and / or blocks.

[0192] These computer program instructions can also be stored in a computer-readable storage medium that can guide the computer or other programmable data processing terminal device to work in a specific manner, so that the instructions stored in the computer-readable storage medium produce a manufactured product including instruction devices that implement the flow Figure 1 The flow or the plurality of flows and / or blocks Figure 1the functions specified in the individual block or blocks. Such computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable devices to generate a computer-implemented process, thus the instructions executed on the computer or other programmable devices provide a process for implementing the functions specified in the flowchart block(s) or block(s). Figure 1 the functions specified in the individual block or blocks. Such computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable devices to generate a computer-implemented process, thus the instructions executed on the computer or other programmable devices provide a process for implementing the functions specified in the flowchart block(s) or block(s). Figure 1 the functions specified in the individual block or blocks. Such computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable devices to generate a computer-implemented process, thus the instructions executed on the computer or other programmable devices provide a process for implementing the functions specified in the flowchart block(s) or block(s).

[0193] It should also be noted that, in the present document, the terms such as first and second, etc. are merely used to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "include", "contain" 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 not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or terminal device. Without more limitations, the element defined by the statement "including a…", does not exclude the presence of other identical elements in the process, method, article or terminal device including the element. In addition, the term "and / or" is merely a description of the association relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the existence of A alone, the existence of A and B together, and the existence of B alone, where A and B can be singular or plural. In addition, the character " / " in the present document generally represents an "or" relationship between the preceding and following associated objects, but it can also represent an "and / or" relationship, which can be understood in the context before and after. "At least one" means one or more, and "multiple" means two or more. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can represent a, b, c, a-b, a-c, b-c or a-b-c, where a, b and c can be single or multiple.

[0194] In addition, it can be understood that in various embodiments of the present application, the size of the sequence number of the above processes does not mean the order of execution, and the execution order of the processes should be determined by their functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0195] Those skilled in the art can appreciate that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware or in a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solutions. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0196] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely schematic, for example, the division of functional modules / units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another device, or some features can be omitted or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed units can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms. The units described as separate components can be or can not be physically separated, and the components displayed as units can be or can not be physical units, that is, can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment. In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present, or two or more units can be integrated in one unit.

[0197] If the method is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in the embodiments of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0198] Finally, it should be noted that the above description is only the preferred embodiment of the application, it should be pointed out that although the preferred embodiment of the application has been described, for those skilled in the art, once the basic creative concept of the application is known, several improvements and refinements can be made without departing from the principles of the application, and these improvements and refinements should also be considered as the protection scope of the application. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the application.

Claims

1. A method for diagnosing the defect area of ​​the glenoid fossa and humeral head, characterized in that: include: Acquiring medical imaging data of the glenoid cavity and humeral head to be processed; Inputting the medical image data into a preset segmentation model to obtain segmentation results of the glenoid fossa and the humeral head; Based on the segmentation results, calculating the defect area and defect ratio of the glenoid cavity; Based on the defect ratio, the degree of humeral head defect is determined.

2. The method for diagnosing the defect area of ​​the glenoid fossa and humeral head according to claim 1, wherein: The segmentation model includes multiple feature enhancement modules that form a Transformer structure. Starting from the second feature enhancement module, the output of each feature enhancement module is processed by the adaptive medical convolution module and the edge attention mechanism module respectively, and the resulting feature data is input into the segmentation head. The segmentation head segments the glenoid cavity and the humeral head based on multiple sets of input feature data and outputs the segmentation results of the glenoid cavity and the humeral head.

3. The method for diagnosing the defect area of ​​the glenoid fossa and humeral head according to claim 2, wherein: The feature enhancement module includes an overlapping block embedding module and a plurality of feature enhancement units; wherein the plurality of feature enhancement units are connected in series, and two adjacent feature enhancement units are connected via a downsampling layer; The overlapping block embedding module introduces overlapping patch encoding and transforms the input image into a 7×7 convolution layer. Each vector in the result represents an overlapping region embedding with context, which serves as the token sequence input to the Transformer structure. The input sequence x input to the feature enhancement unit first passes through the feedforward feedback network to obtain the output x′, which is expressed as follows: x′=x+γ·Conv 1×1 (LN(x))+FFN(x) Among them, γ is a learnable scaling parameter; Conv 1×1 It is a local convolution feedback channel used to embed spatial structure information; LN(x) represents the normalization of the input; FFN represents the independent transformation of each Token, and its transformation operation is expressed as: FFN(x)=W2·σ(W1x+b1)+b2 Where W1 and W2 represent weight matrices, σ represents a nonlinear activation function used to normalize the input; b1 and b2 represent bias vectors; x′ then enters the dimensional interaction module to enhance direct information exchange between tokens at different positions. The cascaded group attention mechanism module then divides x′ processed by the dimensional interaction module into G groups along the channel dimension. Local self-attention is then performed independently on each group to reduce computational complexity. Finally, the outputs of each group are connected in series and then an information fusion operation is performed as the output of the feature enhancement unit.

4. The method for diagnosing the defect area of ​​the glenoid fossa and humeral head according to claim 3, wherein: The number of the feature enhancement units is 3; in the first feature enhancement unit, G is set to 16, in the second feature enhancement unit, G is set to 8, and in the last feature enhancement unit, G is set to 4.

5. The method for diagnosing the defect area of ​​the glenoid fossa and humeral head according to claim 3, wherein: The operational formula of the local self-attention is expressed as: Among them, Q (i) represents the i-th Query matrix; K (i) is the i-th Key matrix; V (i) is the i-th Value matrix; d k is the dimension of Key; T represents the transpose of the matrix; Z (i) Represents the output result of the i-th attention head; Softmax( ) represents normalization processing.

6. The method for diagnosing the defect area of ​​the glenoid fossa and humeral head according to claim 2, wherein: The data processing process of the edge attention mechanism module includes: The Sobel operator is used to extract the first-order gradient information of the input image in the horizontal and vertical directions respectively; among them, the horizontal operator G x and the vertical operator G y Expressed as: Each channel of the input image is matched with G x Perform convolution operation to obtain the horizontal gradient map, and convolve each channel of the input image with G y Perform convolution operation to obtain the vertical gradient map; The horizontal gradient map and the vertical gradient map are processed separately by the global average pooling layer to obtain the aggregate tensor T of the input image in the vertical direction. H and the horizontal aggregation tensor T W ; T H and T W Multiply them together to get a position-related attention map M; M is input into a 1×1 convolutional layer, and then the output of the 1×1 convolutional layer is added to the input image of the edge attention mechanism module to obtain the optimized feature map.

7. The method for diagnosing the defect area of ​​the glenoid fossa and humeral head according to claim 2, wherein: The data processing process of the segmentation header includes: Fuse the input multiple sets of feature data to obtain the fused feature data F fused ; Dynamically calculate the weight of each pixel; the weight generation formula is: W=σ(Conv(F fused )) Where W is a dynamically generated weight mask; σ represents a nonlinear activation function used to normalize the input; Conv(·) represents a convolution operation; the range of W is [0, 1], and a larger value indicates that the model pays more attention to the corresponding area; According to W, the final dynamic mask is generated to guide segmentation; the dynamic mask generation formula is: M dynamic =W⊙P Among them, M dynamic is a dynamic mask; P is the initial prediction result of the model; ⊙ represents pixel-by-pixel weighting; Based on the generated dynamic mask, the glenoid fossa and humeral head are segmented.

8. The method for diagnosing the defect area of ​​the glenoid fossa and humeral head according to claim 1, wherein: Based on the segmentation results, the defect area and defect ratio of the glenoid cavity are calculated, including: Using the segmentation model, generating a standard complete shape of the glenoid cavity based on training data of healthy samples; generating an actual shape of the glenoid cavity based on the medical image data using the segmentation model; Calculating the pixel area difference between the standard complete form and the actual form to obtain the defect area of ​​the glenoid cavity; The percentage of the defect area to the complete glenoid area was calculated to obtain the glenoid defect ratio.

9. The method for diagnosing the defect area of ​​the glenoid fossa and humeral head according to claim 1, wherein: Based on the defect ratio, the degree of humeral head defect is determined, including: If the defect ratio is less than 10%, it is determined to be a slight defect; If the defect ratio is between 10% and 30%, it is judged as moderate defect; If the defect ratio is greater than 30%, it is determined to be a severe defect.

10. A diagnostic system for glenoid and humeral head defect area, characterized in that: include: A data acquisition module, used for acquiring medical imaging data of the glenoid fossa and humeral head to be processed; An object segmentation module is used to input the medical image data into a preset segmentation model to obtain segmentation results of the glenoid cavity and the humeral head; a data calculation module, configured to calculate the defect area and defect ratio of the glenoid cavity based on the segmentation result; The defect degree judgment module is used to determine the defect degree of the humeral head based on the defect ratio.

Images