Muscle-bone ultrasound image healthy and affected side comparison similarity determination method and device

Abstract

The invention provides a muscle-bone ultrasound image healthy and affected side comparison similarity determination method and device. The method comprises the following steps: acquiring a healthy side muscle-bone ultrasound image and an affected side muscle-bone ultrasound image sequence; determining a first segmentation mask graph corresponding to the healthy side muscle-bone ultrasound image and a second segmentation mask graph corresponding to each affected side muscle-bone ultrasound image; determining a first topological feature vector corresponding to the first segmentation mask graph and a second topological feature vector corresponding to each second segmentation mask graph; and based on the first segmentation mask image, the second segmentation mask image, the first topological feature vector and the second topological feature vector, comparing the uninjured-side muscle-bone ultrasound image with each affected-side muscle-bone ultrasound image to obtain the similarity between the uninjured-side muscle-bone ultrasound image and each affected-side muscle-bone ultrasound image. According to the method, the similarity between the healthy side muscle-bone ultrasound image and the affected side muscle-bone ultrasound image is determined by segmenting the mask graph and the topological feature vector, the same section of the healthy side and the affected side is determined quickly and efficiently, and the accuracy of an examination result is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method and device for determining the similarity of healthy and affected sides of musculoskeletal ultrasonic images. Background Art

[0002] The healthy-affected-side comparison method is often used in musculoskeletal scanning to diagnose diseases. The key to using the healthy-affected-side comparison method is to determine the same section of the healthy-affected-side. However, in the practice of musculoskeletal ultrasound scanning, due to factors such as patient status, doctor's professional skills and operation techniques, and ultrasound physical properties, it is impossible to accurately obtain the same section of the healthy-affected-side, thus affecting the accuracy of the examination results. Summary of the invention

[0003] The present invention provides a method and device for determining the similarity of healthy-side and affected-side musculoskeletal ultrasound images, so as to solve the defect that the same section plane of the healthy-side and affected-side cannot be accurately obtained in the prior art, which affects the accuracy of the examination result. The similarity between the healthy-side musculoskeletal ultrasound image and the affected-side musculoskeletal ultrasound image is determined by segmentation mask images and topological feature vectors, so as to quickly and efficiently determine the same section plane of the healthy-side and affected-side, and improve the accuracy of the examination result.

[0004] The present invention provides a method for determining the similarity of healthy and affected sides of musculoskeletal ultrasound images, comprising: Acquire a healthy side muscle-bone ultrasound image and an affected side muscle-bone ultrasound image sequence; a first detection target in the healthy side muscle-bone ultrasound image corresponds to a second detection target in the affected side muscle-bone ultrasound image sequence; the affected side muscle-bone ultrasound image sequence is at least one affected side muscle-bone ultrasound image acquired from different sections of the second detection target; Determine a first segmentation mask image corresponding to the healthy-side muscle-bone ultrasound image and a second segmentation mask image corresponding to each of the affected-side muscle-bone ultrasound images; Determine a first topological feature vector corresponding to the first segmentation mask image and a second topological feature vector corresponding to each of the second segmentation mask images; Based on the first segmentation mask image, the second segmentation mask image, the first topological feature vector and the second topological feature vector, the healthy-side musculoskeletal ultrasound image is compared with each of the affected-side musculoskeletal ultrasound images to obtain the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images; the similarity is used to search for the target affected-side musculoskeletal ultrasound image corresponding to the healthy-side musculoskeletal ultrasound image from the affected-side musculoskeletal ultrasound image sequence.

[0005] According to a method for determining the similarity of the healthy-side and affected-side comparison of musculoskeletal ultrasound images provided by the present invention, the method for determining a first segmentation mask image corresponding to the healthy-side musculoskeletal ultrasound image and a second segmentation mask image corresponding to each of the affected-side musculoskeletal ultrasound images comprises: inputting the healthy-side musculoskeletal ultrasound image and the affected-side musculoskeletal ultrasound image into a musculoskeletal ultrasound image classification model to obtain a first category label corresponding to the healthy-side musculoskeletal ultrasound image and a second category label corresponding to each of the affected-side musculoskeletal ultrasound images; the musculoskeletal ultrasound image classification model is trained based on a musculoskeletal ultrasound image dataset and a label set corresponding to the category label of the musculoskeletal ultrasound image dataset; inputting the first category label into a first target segmentation model corresponding to the first category label to obtain the first segmentation mask image. Mask map; the first segmentation mask map is used to characterize different anatomical structures in the healthy side muscle-bone ultrasound image; the first target segmentation model is trained based on the first sample muscle-bone ultrasound image data set corresponding to the first category label and the segmentation mask map label corresponding to the first sample muscle-bone ultrasound image data set; the second category label is input into the second target segmentation model corresponding to the second category label to obtain the second segmentation mask map; the second segmentation mask map is used to characterize different anatomical structures in the affected side muscle-bone ultrasound image; the second target segmentation model is trained based on the second sample muscle-bone ultrasound image data set corresponding to the second category label and the segmentation mask map label corresponding to the second sample muscle-bone ultrasound image data set.

[0006] According to a method for determining the similarity of healthy-side and affected-side comparison of musculoskeletal ultrasound images provided by the present invention, the method for determining the first topological feature vector corresponding to the first segmentation mask image and the second topological feature vector corresponding to each of the second segmentation mask images includes: performing centroid processing on the first segmentation mask image and the second segmentation mask image, respectively, to obtain a first adjacency matrix and a second adjacency matrix; respectively inputting the first adjacency matrix and the second adjacency matrix into a graph convolutional network, to obtain the first topological feature vector and the second topological feature vector.

[0007] According to a method for determining the similarity of healthy-side and affected-side musculoskeletal ultrasound image comparison provided by the present invention, the healthy-side musculoskeletal ultrasound image is compared with each of the affected-side musculoskeletal ultrasound images based on the first segmentation mask image, the second segmentation mask image, the first topological feature vector and the second topological feature vector to obtain the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images, including: based on the pixels of the first segmentation mask image and the pixels of the second segmentation mask image, the healthy-side musculoskeletal ultrasound image is compared with each of the affected-side musculoskeletal ultrasound images to obtain the first similarity between the first segmentation mask image and the second segmentation mask image; determining the second similarity between the first topological feature vector and the second topological feature vector; and fusing the first similarity and the second similarity to obtain the similarity between the healthy-side musculoskeletal ultrasound image and the affected-side musculoskeletal ultrasound image.

[0008] According to a method for determining the similarity of healthy-side and affected-side musculoskeletal ultrasound image comparison provided by the present invention, after the healthy-side musculoskeletal ultrasound image is compared with each of the affected-side musculoskeletal ultrasound images based on the first segmentation mask image, the second segmentation mask image, the first topological feature vector and the second topological feature vector to obtain the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images, the method further includes: sorting the affected-side musculoskeletal ultrasound images based on the similarity to obtain sorted affected-side musculoskeletal ultrasound images; and using the affected-side musculoskeletal ultrasound image whose third similarity corresponding to the sorted affected-side musculoskeletal ultrasound image is greater than a similarity threshold as the target affected-side musculoskeletal ultrasound image corresponding to the healthy-side musculoskeletal ultrasound image.

[0009] According to a method for determining the similarity of healthy-side and affected-side musculoskeletal ultrasound image comparison provided by the present invention, the method of obtaining a healthy-side musculoskeletal ultrasound image and an affected-side musculoskeletal ultrasound image sequence comprises: obtaining an original healthy-side musculoskeletal ultrasound image and at least one original affected-side musculoskeletal ultrasound image; performing image normalization processing and image scaling processing on the original healthy-side musculoskeletal ultrasound image and each of the original affected-side musculoskeletal ultrasound images, respectively, to obtain the healthy-side musculoskeletal ultrasound image and the affected-side musculoskeletal ultrasound image sequence.

[0010] The present invention also provides a device for determining the similarity of healthy and affected sides of musculoskeletal ultrasound images, comprising the following modules: An acquisition module is used to acquire a healthy side muscle-bone ultrasound image and an affected side muscle-bone ultrasound image sequence; a first detection target in the healthy side muscle-bone ultrasound image corresponds to a second detection target in the affected side muscle-bone ultrasound image sequence; and the affected side muscle-bone ultrasound image sequence is at least one affected side muscle-bone ultrasound image acquired from different sections of the second detection target; A first determination module is used to determine a first segmentation mask image corresponding to the healthy side muscle-bone ultrasound image and a second segmentation mask image corresponding to each of the affected side muscle-bone ultrasound images; A second determining module, configured to determine a first topological feature vector corresponding to the first segmentation mask image and a second topological feature vector corresponding to each of the second segmentation mask images; The third determination module is used to compare the healthy-side musculoskeletal ultrasound image with each of the affected-side musculoskeletal ultrasound images based on the first segmentation mask image, the second segmentation mask image, the first topological feature vector and the second topological feature vector to obtain the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images; the similarity is used to search for the target affected-side musculoskeletal ultrasound image corresponding to the healthy-side musculoskeletal ultrasound image from the affected-side musculoskeletal ultrasound image sequence.

[0011] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for determining the similarity of healthy and affected side comparison of musculoskeletal ultrasound images as described in any one of the above is implemented.

[0012] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for determining the similarity of the healthy and affected sides of musculoskeletal ultrasound images as described in any one of the above methods is implemented.

[0013] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-mentioned methods for determining the similarity of healthy and affected side comparison of musculoskeletal ultrasound images.

[0014] The method, device, equipment, medium and product for determining the similarity of the healthy and affected side comparison of musculoskeletal ultrasound images provided by the present invention obtain a sequence of healthy side musculoskeletal ultrasound images and affected side musculoskeletal ultrasound images; determine a first segmentation mask corresponding to the healthy side musculoskeletal ultrasound image and a second segmentation mask corresponding to each affected side musculoskeletal ultrasound image; determine a first topological feature vector corresponding to the first segmentation mask and a second topological feature vector corresponding to each second segmentation mask; based on the first segmentation mask, the second segmentation mask, the first topological feature vector and the second topological feature vector, determine the similarity between the healthy side musculoskeletal ultrasound image and each affected side musculoskeletal ultrasound image. In this way, the segmentation mask of the musculoskeletal ultrasound image and the topological feature vector of the segmentation mask can be used to comprehensively determine the similarity between the healthy side musculoskeletal ultrasound image and the affected side musculoskeletal ultrasound image, quickly and efficiently determine the same section of the healthy and affected sides, thereby improving the accuracy of the examination results. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 It is a flow chart of a method for determining the similarity of healthy-side and affected-side comparison of musculoskeletal ultrasound images provided by the present invention.

[0017] Figure 2 The invention provides a device for determining the similarity of healthy and affected side comparison of musculoskeletal ultrasonic images.

[0018] Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] It should be noted that musculoskeletal ultrasound is a diagnostic technology that uses conventional ultrasound diagnostic equipment to select ultrasound probes of different frequencies according to the depth of the part to be explored, and evaluates the sound and image manifestations of human joints, tendons, ligaments, bones and cartilages, peripheral nerves and other tissues. Taking peripheral nerves as an example, musculoskeletal ultrasound can clearly display the structure of peripheral nerves and their anatomical relationship with surrounding tissues in real time, from multiple angles, and can explore along the distribution of peripheral nerves. It has the advantages of high-resolution imaging, dynamic observation, simple operation, low examination cost, and no radiation. It has been widely used in many departments such as ultrasound, pain, anesthesia, rehabilitation, and traditional Chinese medicine.

[0021] Since there are significant differences in tissue morphology between different patients, while the tissue morphology of both sides of the same patient is relatively similar, the healthy-affected side comparison method is often used in musculoskeletal scanning to diagnose the disease. The healthy-affected side comparison method refers to scanning the same section of the healthy side and the affected side of the patient, measuring the key tissue structures, and comparing the cross-sectional area, length, width, circumference and other physiological characteristic parameters of the two. Among them, the most critical part of using the healthy-affected side comparison method is to determine the same section of the healthy and affected sides. However, in the practice of musculoskeletal ultrasound scanning, due to factors such as the patient's condition, the doctor's professional skills and operating techniques, and the physical properties of ultrasound, it is impossible to accurately obtain the same section of the healthy and affected sides, which affects the accuracy of the examination results.

[0022] In recent years, domestic and foreign scholars have conducted extensive and in-depth research on the automatic acquisition, recognition, classification and automatic measurement of biological parameters of standard ultrasound sections. However, most of the research focuses on the prenatal fetus, liver, heart and other parts, while there are few studies on the automatic recognition of ultrasound sections of the healthy and affected sides and the automatic measurement of parameters of superficial organs such as muscles and bones. Therefore, there is an urgent need for a muscle-bone ultrasound image section detection method for healthy-side and affected-side comparison, which can quickly and efficiently determine the same section on both sides.

[0023] Based on the above problems, the present invention provides a method for determining the similarity of the healthy-side and affected-side musculoskeletal ultrasound images. The similarity between the healthy-side musculoskeletal ultrasound image and the affected-side musculoskeletal ultrasound image is comprehensively determined by utilizing the segmentation mask map of the musculoskeletal ultrasound image and the topological feature vector of the segmentation mask map, so as to quickly and efficiently determine the same section of the healthy-side and affected-side, thereby improving the accuracy of the examination results.

[0024] Combine the following Figure 1 The present invention describes a method for determining the similarity of a healthy-side and affected-side comparison of musculoskeletal ultrasound images. The method is applicable to determining the similarity of any musculoskeletal ultrasound image. The execution subject of the method may be an electronic device, or may be a method for determining the similarity of a healthy-side and affected-side comparison of musculoskeletal ultrasound images arranged in the electronic device. The device for determining the similarity of a healthy-side and affected-side comparison of musculoskeletal ultrasound images may be implemented by software, hardware, or a combination of both.

[0025] Figure 1 FIG. 1 is a flow chart of a method for determining the similarity of healthy and affected side comparison of musculoskeletal ultrasound images provided by the present invention. Figure 1 As shown, the method includes the following: Step 101: Acquire a healthy side muscle-bone ultrasound image and an affected side muscle-bone ultrasound image sequence.

[0026] Among them, the first detection target in the healthy side muscle-bone ultrasound image corresponds to the second detection target in the affected side muscle-bone ultrasound image sequence; the affected side muscle-bone ultrasound image sequence is at least one affected side muscle-bone ultrasound image collected from different sections of the second detection target.

[0027] Here, the musculoskeletal ultrasound image may be a cross-sectional ultrasound image of structures such as human joints, tendons, ligaments, bones and cartilages, and peripheral nerves.

[0028] It should be noted that the healthy side and the affected side correspond to each other. For example, if the patient's left wrist joint is damaged and the right wrist joint is normal, the left wrist joint is the healthy side and the right wrist joint is the affected side. If the patient's right knee joint is damaged and the left knee joint is normal, the right knee is the healthy side and the left knee is the affected side.

[0029] In one embodiment, the method for obtaining the healthy-side muscle-bone ultrasound image and the affected-side muscle-bone ultrasound image sequence may be any suitable method, for example, obtaining the healthy-side muscle-bone ultrasound image from the Internet, or obtaining the healthy-side muscle-bone ultrasound image by scanning, and then still using the scanning method to continuously adjust the angle and section to obtain multiple affected-side muscle-bone ultrasound images, that is, the affected-side muscle-bone ultrasound image sequence.

[0030] Exemplarily, taking the carpal tunnel section at the proximal end of the wrist joint as an example, the section is divided into the carpal tunnel structure and the carpal tunnel structure. Carpal tunnel: The transverse carpal ligament and the carpal groove composed of 8 carpal bones together form a bone fiber tunnel, and there are 9 flexor tendons and 1 nerve in the tunnel. The 9 flexor tendons are the flexor pollicis longus tendon (1), the superficial flexor tendons of the digitorum (4), and the deep flexor tendons of the digitorum (4). The median nerve becomes flat in the carpal tunnel and is close to the deep radial surface of the flexor retinaculum. Structure outside the carpal tunnel: There are three tendons, namely the ulnar flexor tendon of the wrist, the radial flexor tendon of the wrist, and the palmaris longus tendon. In addition, there is the ulnar carpal tunnel. Ulnar carpal tunnel. The ulnar carpal tunnel is a triangular structure, with the pisiform bone on the ulnar side, the transverse carpal ligament at the bottom, and the retinaculum at the top, and the ulnar artery, ulnar vein and ulnar nerve inside.

[0031] Taking the section of the proximal carpal tunnel of the wrist as an example, when scanning, the probe is placed horizontally on the palm side of the wrist joint to find the scaphoid tubercle (radial side) and pisiform bone (ulnar side) of the proximal carpal tunnel. The lunate bone and triangular bone are the bottom of the carpal tunnel, and the transverse carpal ligament is in front of the carpal tunnel. Inside the carpal tunnel: the shallowest structure is the median nerve, which is close to the transverse carpal ligament, and the short axis is flat round mesh-like low echo. The radial side behind the median nerve is the flexor pollicis longus tendon, which is close to the scaphoid bone. There are four superficial flexor tendons of the fingers (superficial surface) and four deep flexor tendons of the fingers (deep surface) on the deep surface of the median nerve. Outside the carpal tunnel: the radial flexor tendon of the wrist is located on the shallow side of the scaphoid bone, the ulnar flexor tendon of the wrist is located on the shallow side of the pisiform bone, and the palmaris longus tendon is located in the middle. It is very thin and needs to be determined by dynamic scanning. The palmaris longus tendon continues to the palm as the palmar aponeurosis, which is located on the shallow side of the tendon and muscle. Although it is not easy to show under normal circumstances, it is an important anatomical structure. When it is thickened in pathological conditions, it is easy to show by ultrasound. Ulnar canal: It is a bony fibrous canal on the ulnar side of the wrist, through which the ulnar artery, ulnar vein, and ulnar nerve pass.

[0032] In the embodiment of the present invention, the healthy side musculoskeletal ultrasound image and the affected side musculoskeletal ultrasound image sequence may be unprocessed images or preprocessed images. The following is the process of preprocessing the musculoskeletal ultrasound images.

[0033] The method of acquiring a healthy-side musculoskeletal ultrasound image and an affected-side musculoskeletal ultrasound image sequence comprises: acquiring an original healthy-side musculoskeletal ultrasound image and at least one original affected-side musculoskeletal ultrasound image; and performing image normalization processing and image scaling processing on the original healthy-side musculoskeletal ultrasound image and each of the original affected-side musculoskeletal ultrasound images, respectively, to obtain the healthy-side musculoskeletal ultrasound image and the affected-side musculoskeletal ultrasound image sequence.

[0034] Here, image normalization processing refers to adjusting the pixel values ​​of an image to a specific range, wherein the methods of image normalization processing include but are not limited to minimum-maximum normalization, Z-Score Normalization, image histogram equalization, etc.

[0035] Image scaling refers to scaling the musculoskeletal ultrasound image to a fixed size. Generally, the fixed size is 256. 256.

[0036] In an embodiment of the present invention, each muscle-bone ultrasound image is normalized to eliminate brightness differences and contrast differences in different muscle-bone ultrasound images, so that image processing can be more stable and efficient. Image scaling is also performed on each muscle-bone image so that the image can meet the requirements of different models for input images, thereby improving the efficiency and effect of image processing.

[0037] Step 102: determine a first segmentation mask image corresponding to the healthy-side muscle-bone ultrasound image and a second segmentation mask image corresponding to each of the affected-side muscle-bone ultrasound images.

[0038] Here, the segmentation mask image is used to characterize different anatomical structure regions in the musculoskeletal ultrasound image, wherein the segmentation mask image may include the key anatomical structure name, category, and confidence level of each pixel in the musculoskeletal ultrasound image.

[0039] Further, the determining of the first segmentation mask image corresponding to the healthy-side musculoskeletal ultrasound image and the second segmentation mask image corresponding to each of the affected-side musculoskeletal ultrasound images includes: inputting the healthy-side musculoskeletal ultrasound image and the affected-side musculoskeletal ultrasound image into a musculoskeletal ultrasound image classification model to obtain a first category label corresponding to the healthy-side musculoskeletal ultrasound image and a second category label corresponding to each of the affected-side musculoskeletal ultrasound images; the musculoskeletal ultrasound image classification model is trained based on a musculoskeletal ultrasound image dataset and a label set of category labels corresponding to the musculoskeletal ultrasound image dataset; inputting the first category label into a first target segmentation model corresponding to the first category label to obtain the first segmentation mask image; the first segmentation mask image The figure is used to characterize the different anatomical structures in the healthy side musculoskeletal ultrasound image; the first target segmentation model is trained based on the first sample musculoskeletal ultrasound image data set corresponding to the first category label and the segmentation mask map label corresponding to the first sample musculoskeletal ultrasound image data set; the second category label is input into the second target segmentation model corresponding to the second category label to obtain the second segmentation mask map; the second segmentation mask map is used to characterize the different anatomical structures in the affected side musculoskeletal ultrasound image; the second target segmentation model is trained based on the second sample musculoskeletal ultrasound image data set corresponding to the second category label and the segmentation mask map label corresponding to the second sample musculoskeletal ultrasound image data set.

[0040] Here, the musculoskeletal ultrasound image classification model may be trained based on a musculoskeletal ultrasound image dataset carrying classification annotation information; the category label includes a section name.

[0041] Here, the musculoskeletal ultrasound image classification model may be any appropriate model, for example, a classification model based on YOLOv8 (YouOnly Look Once Version 8), a support vector model, a random forest model, etc.

[0042] Here, the category label is also called the section name or tissue. Taking the wrist joint as an example, the sections can be the cross section of the 1st chamber, the cross section of the 2nd-5th chamber, the distal cross section of the 4th chamber, the cross section of the distal radioulnar joint, the longitudinal section of the wrist joint, the proximal cross section of the 6th chamber, the longitudinal section of the 6th chamber, the cross section of the ulnar styloid process, the longitudinal section of the ulnar styloid process, the dorsal side of the 1st metacarpophalangeal joint, the dorsal side of the 2nd metacarpophalangeal joint, the dorsal side of the 2nd proximal interphalangeal joint, the cross section of the proximal carpal tunnel, the cross section of the distal carpal tunnel, the palmar side of the 1st metacarpophalangeal joint, the palmar side of the 2nd proximal interphalangeal joint, the cross section of the flexor pollicis longus tendon, the cross section of the 2nd / 3rd finger flexor tendon, the long axis of the flexor pollicis longus tendon (1st metacarpophalangeal joint) and the longitudinal section of the carpal tunnel.

[0043] Exemplarily, the musculoskeletal ultrasound image classification model may be a trained YOLOv8-based classification model, which may be trained based on the above-mentioned musculoskeletal ultrasound image dataset, and the annotation information may include the section names of each target result in the musculoskeletal ultrasound image annotated by a senior doctor, etc. The first side musculoskeletal ultrasound image and the second musculoskeletal ultrasound image sequence are input into the trained YOLOv8-based classification model, and the section name described in the image can be output, that is, the category label corresponding to the image is obtained.

[0044] The classification model based on YOLOv8 mainly consists of three parts: backbone network (Backbone), feature enhancement network (Neck) and detection head (Head).

[0045] Backbone downsamples the input ultrasound image, extracts the features of key anatomical structures through continuous convolution operations, and generates corresponding feature maps. Backbone refers to the Cross Stage PartialDarkNet (CSPDarkNet) structure and uses the C2F module. The C2F module has fewer parameters and better feature extraction capabilities, which helps to make the network lightweight. In addition, YOLOv8 also retains the Spatial PyramidPooling Fast (SPPF) module to enhance feature representation capabilities.

[0046] Neck adopts the idea of ​​Pixel Aggregation Network-Feature Pyramid Network (PAN-FPN), but optimizes its structure. Specifically, it removes the convolution structure of the PAN-FPN upsampling stage in YOLOv5 and replaces the C3 module with the C2 module, making its feature fusion capability more efficient and improving model performance. It re-upsamples the downsampled feature map for feature fusion to adapt to features of different sizes of various anatomical structures.

[0047] In the Head, the decoupled-head idea is adopted to separate the regression branch and the classification branch. This design makes the training and reasoning of the model network faster. At the same time, YOLOv8 abandons the anchor-based solution and adopts the anchor-free idea to further simplify the model structure.

[0048] It should be noted that for each category label, a trained target segmentation model is preset. For example, for the proximal carpal tunnel section, a proximal carpal tunnel section target segmentation model for predicting the proximal carpal tunnel section can be preset. For the long axis or short axis section of the quadriceps tendon, a long axis or short axis target segmentation model of the quadriceps tendon can be preset for predicting the long axis or short axis of the quadriceps tendon. After obtaining the first side muscle-bone ultrasound image or the second side muscle-bone ultrasound image sequence, the corresponding target segmentation network is called for prediction according to the category label to obtain the corresponding segmentation mask map. Specifically, the segmentation mask map may include the key anatomical structure name, category, and confidence of each pixel of the muscle-bone ultrasound image.

[0049] In an embodiment of the present invention, the target segmentation model may include the following network structures: U-Net, DM-ResNet, Conformer, FAT-Net and DA-TransUNet. The target segmentation models of the above five network architectures are respectively packaged. When using the method of this implementation, one of the above five target segmentation models is selected for calculation according to actual needs. When the calculation speed is given priority, U-Net and DM-ResNet are selected. When the segmentation accuracy is limited, Conformer and FAT-Net are selected. When the calculation speed and segmentation accuracy are comprehensively considered, DA-TransUNet is selected. For the target segmentation of musculoskeletal ultrasound images, five independent algorithms are encapsulated, so that a more suitable algorithm can be selected quickly and conveniently according to actual needs.

[0050] In an embodiment of the present invention, the musculoskeletal ultrasound image is classified and segmented in turn by using a musculoskeletal ultrasound image classification model and a target segmentation model to obtain different key anatomical structures in the musculoskeletal ultrasound image. The similarity between the healthy side musculoskeletal ultrasound image and the affected side musculoskeletal ultrasound image is fully determined based on the anatomical structure, the accuracy of the similarity is improved, and the same section on both sides can be determined quickly and efficiently.

[0051] Step 103: Determine a first topological feature vector corresponding to the first segmentation mask image and a second topological feature vector corresponding to each of the second segmentation mask images.

[0052] Here, the topological feature vector is usually extracted from the original data, which can reflect the characteristics of the data in the topological structure, including the connectivity of the data, the importance of the nodes, the weight of the edges, etc.

[0053] Here, the method for determining the topological eigenvector may be any appropriate method, such as a method based on graph theory, a method based on continuous homology, a method based on complex network analysis, and the like.

[0054] Exemplarily, determining the first topological feature vector corresponding to the first segmentation mask map and the second topological feature vector corresponding to each of the second segmentation mask maps includes: performing centroid processing on the first segmentation mask map and the second segmentation mask map, respectively, to obtain a first adjacency matrix and a second adjacency matrix; respectively inputting the first adjacency matrix and the second adjacency matrix into a graph convolutional network to obtain the first topological feature vector and the second topological feature vector.

[0055] Here, the centroiding process actually determines the centroids of different anatomical structures in the segmentation mask. Then it is used as the nodes of the graph, and the lines connecting two adjacent nodes are used as the edges of the graph, and the topological graph G=(V, E) is obtained, where V and E represent the nodes and a set of edges in the graph respectively. For multiple node topological graphs, the feature matrix can be used , the feature matrix has k nodes, each node has a d-dimensional feature vector. Taking the proximal carpal tunnel of the wrist joint as an example, the selected key tissues are the median nerve, the deep flexor tendon of the digitorum, the superficial flexor tendon of the digitorum, the radial artery, the flexor tendon of the pollicis longus, the flexor tendon of the palmaris longus and the transverse carpal ligament, so k is set to 7. Indicates the size of the segmentation map, equal Multiply .

[0056] According to the topological graph, construct its adjacency matrix , the adjacency matrix is ​​expressed as follows (1):

[0057] in, Represents an edge.

[0058] Further, in obtaining and After that, it is input into the multi-layer graph convolutional network (GCN), and the calculation method is as follows formula (2): (2) in, represents the feature matrix after activation, represents the activation function, yes The degree matrix of represents the sum of the adjacency matrix and the identity matrix, is a trainable weight matrix.

[0059] For example, The following formula (3): = + (3) in, Represents the identity matrix.

[0060] Furthermore, the output of the graph convolutional network is as follows: (4) in, represents the network output topology feature vector, and is the weight matrix of linear projection, and Represent the intermediate dimension and the final number of classes respectively.

[0061] In the embodiment of the present invention, by converting the segmentation mask graph into an adjacency matrix, valuable topological information is encoded into an adjacency matrix, and information that is critical to the system topology is retained, thereby avoiding the loss of key features and improving the accuracy of similarity.

[0062] Step 104: Based on the first segmentation mask image, the second segmentation mask image, the first topological feature vector and the second topological feature vector, the healthy-side musculoskeletal ultrasound image is compared with each of the affected-side musculoskeletal ultrasound images to obtain the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images.

[0063] It should be noted that the similarity is used to characterize the difference between the healthy side muscle-bone ultrasound image and the affected side muscle-bone ultrasound image, wherein the higher the similarity, the smaller the difference; the lower the similarity, the greater the difference.

[0064] The similarity is used to search the target affected-side musculoskeletal ultrasound image corresponding to the healthy-side musculoskeletal ultrasound image from the affected-side musculoskeletal ultrasound image sequence.

[0065] Furthermore, based on the first segmentation mask map, the second segmentation mask map, the first topological feature vector and the second topological feature vector, the healthy-side musculoskeletal ultrasound image is compared with each of the affected-side musculoskeletal ultrasound images to obtain the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images, including: based on the pixels of the first segmentation mask map and the pixels of the second segmentation mask map, the healthy-side musculoskeletal ultrasound image is compared with each of the affected-side musculoskeletal ultrasound images to obtain the first similarity between the first segmentation mask map and the second segmentation mask map; determining the second similarity between the first topological feature vector and the second topological feature vector; and fusing the first similarity and the second similarity to obtain the similarity between the healthy-side musculoskeletal ultrasound image and the affected-side musculoskeletal ultrasound image.

[0066] Here, methods for determining the first similarity include but are not limited to intersection-over-union ratio, pixel accuracy, Dice coefficient, Hausdorff distance, and the like.

[0067] Here, the method of determining the second similarity includes but is not limited to cosine similarity, Euclidean distance, and Manhattan distance.

[0068] Here, the fusion process may be to add the first similarity and the second similarity; or to perform a weighted sum of the two; or to perform a weighted average of the two, which is not limited in the present invention.

[0069] In the embodiment of the present invention, the similarity between the healthy side muscle-bone ultrasound image and the affected side muscle-bone ultrasound image is comprehensively determined by fusing the first similarity of the segmentation graph and the second similarity of the topological feature vector, thereby improving the accuracy and reliability of the similarity.

[0070] Further, after comparing the healthy-side musculoskeletal ultrasound image with each of the affected-side musculoskeletal ultrasound images based on the first segmentation mask map, the second segmentation mask map, the first topological feature vector and the second topological feature vector to obtain the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images, the method also includes: sorting the affected-side musculoskeletal ultrasound images based on the similarity to obtain sorted affected-side musculoskeletal ultrasound images; and using the affected-side musculoskeletal ultrasound image whose third similarity corresponding to the sorted affected-side musculoskeletal ultrasound image is greater than a similarity threshold as the target affected-side musculoskeletal ultrasound image corresponding to the healthy-side musculoskeletal ultrasound image.

[0071] Here, the target affected-side musculoskeletal ultrasound image is an image that is similar or identical to the detected musculoskeletal ultrasound image.

[0072] It should be noted that the affected-side musculoskeletal ultrasound images can be sorted from large to small according to similarity, or from small to large according to similarity. If the sorting is from small to large, the last N affected-side musculoskeletal ultrasound images are retained; if the sorting is from large to small, the first N affected-side musculoskeletal ultrasound images are retained.

[0073] It should be noted that when the similarity is greater than the similarity threshold, it means that the healthy side musculoskeletal ultrasound image and the affected side musculoskeletal ultrasound image are images of the same cut plane; otherwise, the healthy side musculoskeletal ultrasound image and the affected side musculoskeletal ultrasound image are not images of the same cut plane.

[0074] In an embodiment of the present invention, the target affected-side musculoskeletal ultrasound image corresponding to the healthy-side musculoskeletal ultrasound image is further determined through high-accuracy similarity, thereby improving the accuracy of matching the healthy-side musculoskeletal ultrasound image with the affected-side musculoskeletal ultrasound image, and further improving the accuracy of the examination results.

[0075] In an embodiment of the present invention, the similarity between the healthy-side musculoskeletal ultrasound image and the affected-side musculoskeletal ultrasound image is comprehensively determined by utilizing the segmentation mask map of the musculoskeletal ultrasound image and the topological feature vector of the segmentation mask map, so as to quickly and efficiently determine the same section plane of the healthy and affected sides, thereby improving the accuracy of the examination results.

[0076] The following is a description of the device for determining the similarity between the healthy and affected sides of musculoskeletal ultrasound images provided by the present invention. The device for determining the similarity between the healthy and affected sides of musculoskeletal ultrasound images described below and the method for determining the similarity between the healthy and affected sides of musculoskeletal ultrasound images described above can be referenced to each other.

[0077] Figure 2 The present invention provides a device for determining the similarity between the healthy and affected sides of musculoskeletal ultrasound images. Figure 2 As shown, the apparatus 200 for determining the similarity between the healthy and affected sides of the musculoskeletal ultrasound images comprises: An acquisition module 201 is used to acquire a healthy side muscle-bone ultrasound image and an affected side muscle-bone ultrasound image sequence; a first detection target in the healthy side muscle-bone ultrasound image corresponds to a second detection target in the affected side muscle-bone ultrasound image sequence; and the affected side muscle-bone ultrasound image sequence is at least one affected side muscle-bone ultrasound image acquired from different sections of the second detection target; A first determination module 202 is used to determine a first segmentation mask image corresponding to the healthy side muscle-bone ultrasound image and a second segmentation mask image corresponding to each of the affected side muscle-bone ultrasound images; A second determination module 203, configured to determine a first topological feature vector corresponding to the first segmentation mask image and a second topological feature vector corresponding to each of the second segmentation mask images; The third determination module 204 is used to compare the healthy-side musculoskeletal ultrasound image with each of the affected-side musculoskeletal ultrasound images based on the first segmentation mask image, the second segmentation mask image, the first topological feature vector and the second topological feature vector to obtain the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images; the similarity is used to search for the target affected-side musculoskeletal ultrasound image corresponding to the healthy-side musculoskeletal ultrasound image from the affected-side musculoskeletal ultrasound image sequence.

[0078] In some embodiments, the first determination module 202 is specifically used to: input the healthy side musculoskeletal ultrasound image and the affected side musculoskeletal ultrasound image into a musculoskeletal ultrasound image classification model to obtain a first category label corresponding to the healthy side musculoskeletal ultrasound image and a second category label corresponding to each affected side musculoskeletal ultrasound image; the musculoskeletal ultrasound image classification model is trained based on a musculoskeletal ultrasound image dataset and a label set of category labels corresponding to the musculoskeletal ultrasound image dataset; input the first category label into a first target segmentation model corresponding to the first category label to obtain the first segmentation mask map; the first segmentation mask map is used to characterize different anatomical structures in the healthy side musculoskeletal ultrasound image; the first target segmentation model is trained based on a first sample musculoskeletal ultrasound image dataset corresponding to the first category label and a segmentation mask map label corresponding to the first sample musculoskeletal ultrasound image dataset; input the second category label into a second target segmentation model corresponding to the second category label to obtain the second segmentation mask map; the second segmentation mask map is used to characterize different anatomical structures in the affected side musculoskeletal ultrasound image; the second target segmentation model is trained based on a second sample musculoskeletal ultrasound image dataset corresponding to the second category label and a segmentation mask map label corresponding to the second sample musculoskeletal ultrasound image dataset.

[0079] In some embodiments, the second determination module 203 is specifically used to: perform centroid processing on the first segmentation mask map and the second segmentation mask map, respectively, to obtain a first adjacency matrix and a second adjacency matrix; input the first adjacency matrix and the second adjacency matrix into a graph convolutional network, respectively, to obtain the first topological feature vector and the second topological feature vector.

[0080] In some embodiments, the third determination module 204 is specifically used to: compare the healthy side muscle-bone ultrasound image with each of the affected side muscle-bone ultrasound images based on the pixels of the first segmentation mask image and the pixels of the second segmentation mask image to obtain a first similarity between the first segmentation mask image and the second segmentation mask image; determine a second similarity between the first topological feature vector and the second topological feature vector; and fuse the first similarity and the second similarity to obtain the similarity between the healthy side muscle-bone ultrasound image and the affected side muscle-bone ultrasound image.

[0081] In some embodiments, after determining the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images based on the first segmentation mask map, the second segmentation mask map, the first topological feature vector and the second topological feature vector, the healthy-side and affected-side musculoskeletal ultrasound image similarity determination device 200 also includes a search module, which is specifically used to: sort the affected-side musculoskeletal ultrasound images based on the similarity to obtain sorted affected-side musculoskeletal ultrasound images; and use the affected-side musculoskeletal ultrasound image whose third similarity corresponding to the sorted affected-side musculoskeletal ultrasound image is greater than a similarity threshold as the target affected-side musculoskeletal ultrasound image corresponding to the healthy-side musculoskeletal ultrasound image.

[0082] In some embodiments, the acquisition module 201 is specifically used to: acquire an original healthy-side muscle-bone ultrasound image and at least one original affected-side muscle-bone ultrasound image; perform image normalization processing and image scaling processing on the original healthy-side muscle-bone ultrasound image and each of the original affected-side muscle-bone ultrasound images, respectively, to obtain the healthy-side muscle-bone ultrasound image and the affected-side muscle-bone ultrasound image sequence.

[0083] Figure 3 The structural diagram of the electronic device provided by the present invention is as follows: Figure 3 As shown, the electronic device may include: a processor 310 , a communications interface 320 , a memory 330 and a communication bus 340 , wherein the processor 310 , the communications interface 320 , and the memory 330 communicate with each other via the communication bus 340 . The processor 310 can call the logic instructions in the memory 330 to execute a method for determining the similarity of the healthy-side and affected-side comparison of musculoskeletal ultrasound images, the method comprising: obtaining a healthy-side musculoskeletal ultrasound image and an affected-side musculoskeletal ultrasound image sequence; the first detection target in the healthy-side musculoskeletal ultrasound image corresponds to the second detection target in the affected-side musculoskeletal ultrasound image sequence; the affected-side musculoskeletal ultrasound image sequence is at least one affected-side musculoskeletal ultrasound image acquired from different sections of the second detection target; determining a first segmentation mask image corresponding to the healthy-side musculoskeletal ultrasound image and a second segmentation mask image corresponding to each of the affected-side musculoskeletal ultrasound images; determining a first topological feature vector corresponding to the first segmentation mask image and a second topological feature vector corresponding to each of the second segmentation mask images; based on the first segmentation mask image, the second segmentation mask image, the first topological feature vector and the second topological feature vector, determining the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images; the similarity is used to search for a target affected-side musculoskeletal ultrasound image corresponding to the healthy-side musculoskeletal ultrasound image from the affected-side musculoskeletal ultrasound image sequence.

[0084] In addition, the logic instructions in the above-mentioned memory 330 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0085] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the method for determining the similarity of the healthy and affected side comparison of musculoskeletal ultrasound images provided by the above-mentioned methods. The method includes: acquiring a healthy side musculoskeletal ultrasound image and an affected side musculoskeletal ultrasound image sequence; the first detection target in the healthy side musculoskeletal ultrasound image corresponds to the second detection target in the affected side musculoskeletal ultrasound image sequence; the affected side musculoskeletal ultrasound image sequence is at least one affected side musculoskeletal ultrasound image acquired from different sections of the second detection target. bone ultrasound image; determining a first segmentation mask map corresponding to the healthy-side muscle-bone ultrasound image and a second segmentation mask map corresponding to each of the affected-side muscle-bone ultrasound images; determining a first topological feature vector corresponding to the first segmentation mask map and a second topological feature vector corresponding to each of the second segmentation mask maps; determining a similarity between the healthy-side muscle-bone ultrasound image and each of the affected-side muscle-bone ultrasound images based on the first segmentation mask map, the second segmentation mask map, the first topological feature vector and the second topological feature vector; the similarity is used to find a target affected-side muscle-bone ultrasound image corresponding to the healthy-side muscle-bone ultrasound image in the affected-side muscle-bone ultrasound image sequence.

[0086] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the method for determining the similarity of the healthy-side and affected-side musculoskeletal ultrasound images provided by the above-mentioned methods, the method comprising: acquiring a healthy-side musculoskeletal ultrasound image and an affected-side musculoskeletal ultrasound image sequence; a first detection target in the healthy-side musculoskeletal ultrasound image corresponds to a second detection target in the affected-side musculoskeletal ultrasound image sequence; the affected-side musculoskeletal ultrasound image sequence is at least one affected-side musculoskeletal ultrasound image collected from different sections of the second detection target; determining the healthy-side musculoskeletal ultrasound image A first segmentation mask map corresponding to the ultrasound image and a second segmentation mask map corresponding to each of the affected-side muscle-bone ultrasound images; determining a first topological feature vector corresponding to the first segmentation mask map and a second topological feature vector corresponding to each of the second segmentation mask maps; based on the first segmentation mask map, the second segmentation mask map, the first topological feature vector and the second topological feature vector, determining the similarity between the healthy-side muscle-bone ultrasound image and each of the affected-side muscle-bone ultrasound images; the similarity is used to search for a target affected-side muscle-bone ultrasound image corresponding to the healthy-side muscle-bone ultrasound image from the affected-side muscle-bone ultrasound image sequence.

[0087] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0088] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for determining the similarity of healthy and affected side comparison of musculoskeletal ultrasound images, characterized in that: include: Acquire a healthy side muscle-bone ultrasound image and an affected side muscle-bone ultrasound image sequence; a first detection target in the healthy side muscle-bone ultrasound image corresponds to a second detection target in the affected side muscle-bone ultrasound image sequence; the affected side muscle-bone ultrasound image sequence is at least one affected side muscle-bone ultrasound image acquired from different sections of the second detection target; Determine a first segmentation mask image corresponding to the healthy-side muscle-bone ultrasound image and a second segmentation mask image corresponding to each of the affected-side muscle-bone ultrasound images; Determine a first topological feature vector corresponding to the first segmentation mask image and a second topological feature vector corresponding to each of the second segmentation mask images; Based on the first segmentation mask image, the second segmentation mask image, the first topological feature vector and the second topological feature vector, the healthy-side musculoskeletal ultrasound image is compared with each of the affected-side musculoskeletal ultrasound images to obtain the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images; the similarity is used to search for the target affected-side musculoskeletal ultrasound image corresponding to the healthy-side musculoskeletal ultrasound image from the affected-side musculoskeletal ultrasound image sequence.

2. The method for determining the similarity of healthy and affected side comparison of musculoskeletal ultrasound images according to claim 1, characterized in that: The determining of the first segmentation mask image corresponding to the healthy side muscle-bone ultrasound image and the second segmentation mask image corresponding to each of the affected side muscle-bone ultrasound images includes: The healthy side musculoskeletal ultrasound image and the affected side musculoskeletal ultrasound image are input into a musculoskeletal ultrasound image classification model to obtain a first category label corresponding to the healthy side musculoskeletal ultrasound image and a second category label corresponding to each of the affected side musculoskeletal ultrasound images; the musculoskeletal ultrasound image classification model is trained based on a musculoskeletal ultrasound image dataset and a label set of category labels corresponding to the musculoskeletal ultrasound image dataset; The first category label is input into the first target segmentation model corresponding to the first category label to obtain the first segmentation mask map; the first segmentation mask map is used to characterize different anatomical structures in the healthy side muscle-bone ultrasound image; the first target segmentation model is trained based on the first sample muscle-bone ultrasound image data set corresponding to the first category label and the segmentation mask map label corresponding to the first sample muscle-bone ultrasound image data set; The second category label is input into the second target segmentation model corresponding to the second category label to obtain the second segmentation mask map; the second segmentation mask map is used to characterize different anatomical structures in the affected side musculoskeletal ultrasound image; the second target segmentation model is trained based on the second sample musculoskeletal ultrasound image data set corresponding to the second category label and the segmentation mask map label corresponding to the second sample musculoskeletal ultrasound image data set.

3. The method for determining the similarity of healthy and affected side comparison of musculoskeletal ultrasound images according to claim 1 or 2, characterized in that: The determining of the first topological feature vector corresponding to the first segmentation mask image and the second topological feature vector corresponding to each of the second segmentation mask images includes: Performing centroid processing on the first segmentation mask graph and the second segmentation mask graph respectively to obtain a first adjacency matrix and a second adjacency matrix; The first adjacency matrix and the second adjacency matrix are respectively input into a graph convolutional network to obtain the first topological feature vector and the second topological feature vector.

4. The method for determining the similarity of healthy and affected side comparison of musculoskeletal ultrasound images according to claim 1 or 2, characterized in that: The comparing the healthy-side musculoskeletal ultrasound image with each of the affected-side musculoskeletal ultrasound images based on the first segmentation mask image, the second segmentation mask image, the first topological feature vector, and the second topological feature vector to obtain the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images includes: Based on the pixels of the first segmentation mask image and the pixels of the second segmentation mask image, the healthy side muscle-bone ultrasound image is compared with each of the affected side muscle-bone ultrasound images to obtain a first similarity between the first segmentation mask image and the second segmentation mask image; determining a second similarity between the first topological feature vector and the second topological feature vector; The first similarity and the second similarity are fused to obtain the similarity between the healthy side muscle-bone ultrasound image and the affected side muscle-bone ultrasound image.

5. The method for determining the similarity between the healthy and affected sides of musculoskeletal ultrasound images according to claim 4, characterized in that: After comparing the healthy-side musculoskeletal ultrasound image with each of the affected-side musculoskeletal ultrasound images based on the first segmentation mask image, the second segmentation mask image, the first topological feature vector, and the second topological feature vector to obtain the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images, the method further includes: Based on the similarity, the affected-side muscle-bone ultrasound images are sorted to obtain sorted affected-side muscle-bone ultrasound images; The affected-side musculoskeletal ultrasound image whose third similarity corresponding to the sorted affected-side musculoskeletal ultrasound image is greater than the similarity threshold is used as the target affected-side musculoskeletal ultrasound image corresponding to the healthy-side musculoskeletal ultrasound image.

6. The method for determining the similarity of healthy and affected side comparison of musculoskeletal ultrasound images according to claim 1 or 2, characterized in that: The step of acquiring a healthy side muscle-bone ultrasound image and an affected side muscle-bone ultrasound image sequence comprises: Acquire an original healthy-side musculoskeletal ultrasound image and at least one original affected-side musculoskeletal ultrasound image; The original healthy-side musculoskeletal ultrasound image and each of the original affected-side musculoskeletal ultrasound images are respectively subjected to image normalization processing and image scaling processing to obtain the healthy-side musculoskeletal ultrasound image and the affected-side musculoskeletal ultrasound image sequence.

7. A device for determining the similarity between healthy and affected sides of musculoskeletal ultrasound images, characterized in that: include: An acquisition module is used to acquire a healthy side muscle-bone ultrasound image and an affected side muscle-bone ultrasound image sequence; a first detection target in the healthy side muscle-bone ultrasound image corresponds to a second detection target in the affected side muscle-bone ultrasound image sequence; and the affected side muscle-bone ultrasound image sequence is at least one affected side muscle-bone ultrasound image acquired from different sections of the second detection target; A first determination module is used to determine a first segmentation mask image corresponding to the healthy side muscle-bone ultrasound image and a second segmentation mask image corresponding to each of the affected side muscle-bone ultrasound images; A second determining module, configured to determine a first topological feature vector corresponding to the first segmentation mask image and a second topological feature vector corresponding to each of the second segmentation mask images; The third determination module is used to compare the healthy-side musculoskeletal ultrasound image with each of the affected-side musculoskeletal ultrasound images based on the first segmentation mask image, the second segmentation mask image, the first topological feature vector and the second topological feature vector to obtain the similarity between the healthy-side musculoskeletal ultrasound image and each of the affected-side musculoskeletal ultrasound images; the similarity is used to search for the target affected-side musculoskeletal ultrasound image corresponding to the healthy-side musculoskeletal ultrasound image from the affected-side musculoskeletal ultrasound image sequence.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for determining the similarity of the healthy and affected sides of the musculoskeletal ultrasound images is implemented as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for determining the similarity of the healthy and affected sides of musculoskeletal ultrasound images as claimed in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for determining the similarity of the healthy and affected sides of musculoskeletal ultrasound images as claimed in any one of claims 1 to 6 is implemented.

Images