A method and system for precise automatic segmentation of knee joint images

Abstract

The application belongs to the field of image processing, and discloses a knee joint image accurate automatic segmentation auxiliary method and system. The network structure is constructed, and the network structure is optimized according to the required result, so that the network structure is more matched with the knee joint image accurate automatic segmentation demand. Under the same data condition, more effective feature expression and more consistent segmentation output are obtained, and the dependence on single data distribution is reduced. The application obtains the source domain data set for training by processing the preprocessed image data, and trains the optimized network structure by using the source domain data set, so that the training data organization and the training process are more standardized and controllable, and the model is more likely to learn stable segmentation related features, thereby improving the adaptability and generalization ability to different source data, and solving the problems of insufficient model reliability under the conditions of insufficient adaptation of multi-center heterogeneous data and limited scale of high-quality labeling.

Description

Technical Field

[0001] This invention belongs to the field of image processing, specifically relating to a method and system for accurate automatic segmentation of knee joint images. Background Technology

[0002] The knee joint is the largest and most complex weight-bearing joint in the human body. Its intricate anatomy and diverse range of movements make it highly susceptible to injury during daily activities and exercise. Among the many structures in the knee joint, the meniscus and anterior cruciate ligament (ACL) play crucial roles in maintaining joint stability, cushioning loads, and coordinating movement. The meniscus, located between the femur and tibia, is a typical C-shaped fibrocartilage structure. Damage to the meniscus not only significantly weakens the mechanical stability of the knee joint but can also trigger abnormal stress on the articular cartilage, accelerating and irreversibly advancing knee degenerative diseases, severely impacting patients' motor function and quality of life. The ACL, as a core structure limiting anterior tibial displacement and maintaining anteroposterior stability of the knee joint, is considered a significant risk factor for post-traumatic osteoarthritis and significantly increases the long-term risk of requiring total knee replacement surgery. Therefore, accurate and early diagnosis of meniscus and ACL injuries is a prerequisite for precise, stepwise, and personalized treatment of knee joint diseases.

[0003] Magnetic Resonance Imaging (MRI), with its superior soft tissue resolution, non-ionizing radiation, multi-parameter imaging, and multiplanar reconstruction, has become the "gold standard" imaging method for assessing knee injuries, especially meniscus and ACL tears. Numerous studies have demonstrated that MRI has high sensitivity and specificity in diagnosing meniscus tears and has largely replaced diagnostic arthroscopy in clinical practice. However, with the widespread availability of MRI equipment and the continuous increase in examination volume, radiologists are facing unprecedented workload pressure. A complete knee MRI examination typically involves hundreds of multi-sequence, multi-planar images, containing complex and detailed information. Physicians require extensive professional training to accurately diagnose these images, and the interpretation process is time-consuming and labor-intensive. Furthermore, the relative scarcity of professional radiologists in grassroots and remote areas further restricts the accessibility and balance of high-quality knee MRI diagnostic services.

[0004] Against this backdrop, computer-aided diagnostic methods based on artificial intelligence, especially deep learning, have gradually become an important technological direction for solving the aforementioned problems. Deep learning models can automatically learn multi-level, high-dimensional discriminative features from medical image data, establishing a complex mapping relationship from raw images to disease diagnoses end-to-end, demonstrating significant advantages in the field of medical image analysis. For meniscus and ACL injuries, existing research has explored automatic tissue segmentation, damage detection, and graded diagnosis based on deep learning architectures such as U-Net and Mask R-CNN, achieving relatively ideal performance on specific datasets, with some models reaching high levels of diagnostic accuracy. Simultaneously, some studies have introduced interpretability mechanisms, visually presenting the injury areas of interest to the model in the form of heatmaps, thereby improving the transparency and credibility of diagnostic results to some extent.

[0005] Despite positive progress in related research, existing deep learning-based MRI-assisted diagnostic methods for knee joints still face several key challenges in their translation into clinical applications. First, the robustness and generalization ability of the models are generally insufficient. Existing models often rely on data from single centers, specific scanning protocols, and fixed equipment manufacturers for training and validation. When applied to real clinical data from different hospitals, with different equipment, and under different imaging parameters, their diagnostic performance often declines significantly, severely limiting the cross-center promotion and clinical applicability of the models. Second, the scale and quality of training data have become a major bottleneck restricting the improvement of model performance. The datasets used in existing studies generally suffer from limited sample size, inconsistent annotation standards, and imbalanced data distribution. Samples for some injury types are scarce, leading to unstable model learning processes and insufficient consistency in prediction results, making it difficult to meet the stringent requirements of high accuracy and reliability in clinical practice. Summary of the Invention

[0006] The purpose of this invention is to overcome the limitations of existing deep learning-based MRI-assisted diagnostic methods for knee joints, which are generally limited by insufficient adaptability to multi-center heterogeneous data and limited scale of high-quality labeled data, and to provide an auxiliary method and system for accurate automatic segmentation of knee joint images.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for constructing a precise automatic segmentation model for knee joint images, comprising the following steps: Historical knee MRI image data was acquired, and the historical knee MRI image data was preprocessed to obtain preprocessed image data. Build the network structure and optimize it based on the desired results; The preprocessed image data is processed to obtain the source domain dataset for training. The optimized network structure is trained using the source domain dataset to obtain a model for accurate and automatic segmentation of knee joint images.

[0008] A further improvement of this invention lies in the method of acquiring historical knee MRI image data and preprocessing the historical knee MRI image data to obtain preprocessed image data, as follows: Acquire historical knee MRI image data, which includes samples of knee injuries of different types and degrees; Historical knee MRI image data were standardized to obtain preprocessed image data. The standardization process included resampling the spatial resolution of the images, normalizing the grayscale intensity of the images, and suppressing noise and reducing artifacts.

[0009] A further improvement of this invention lies in the following method for standardizing historical knee MRI image data to obtain preprocessed image data: The storage format of historical knee MRI image data was standardized, and the sequence naming and reading methods were standardized to complete the format sequence organization; Resampling was performed on historical knee MRI images with different resolutions and slice thicknesses that had been formatted and processed to a uniform pixel spacing. The grayscale range of the historical knee MRI image data that has been resampled is normalized and upscaled to the required signal-to-noise ratio to obtain the preprocessed image data.

[0010] A further improvement of this invention lies in the following specific method for constructing the network structure and optimizing it according to the desired results: A pre-built Attention U-Net is used as the network structure, and an attention mechanism is introduced into the network structure to complete the construction of the network structure; By integrating a multi-task learning strategy into the network structure, the network structure can learn auxiliary feature recognition tasks related to organization identification while performing organization segmentation, thereby completing network structure optimization.

[0011] A further improvement of the present invention is that the network structure that completes the network structure optimization includes an Attention U-Net backbone network. The decoder of the Attention U-Net backbone network is used to extract multi-scale features from local texture to high-level semantics layer by layer. The decoder is used to restore spatial resolution layer by layer and output segmentation results. Skip connections are used to fuse the detailed information of the encoder and the semantic information of the decoder. An attention mechanism is introduced at the feature fusion location of the encoder and decoder. The attention mechanism is used to weight the features from the encoder and suppress noisy features. The Attention U-Net backbone network employs a multi-task learning strategy, used for segmentation task branches and auxiliary feature recognition task branches.

[0012] A further improvement of this invention lies in processing the preprocessed image data to obtain a source domain dataset for training. The optimized network structure is then trained using the source domain dataset to obtain a precise automatic segmentation model for knee joint images. The specific method is as follows: The preprocessed image data is obtained, and the effective regions in the preprocessed image data are cropped to obtain the target data block containing the knee joint tissue, which is used as the source domain dataset for training. The source domain dataset used for training is divided into K training sets and K test sets using the K-fold cross-validation method. The optimized network structure is trained using K training sets; The trained network structure is validated using K test sets to obtain K optimal models; The key indicators required for the K optimal models are calculated to obtain a model for accurate and automatic segmentation of knee joint images.

[0013] A further improvement of this invention is that, when training the optimized network structure using K training sets, different historical knee joint MRI image data are used as training sets.

[0014] Secondly, the present invention provides a method for constructing a precise automatic segmentation model for knee joint images, comprising: The preprocessing module is used to acquire historical knee MRI image data, preprocess the historical knee MRI image data, and obtain preprocessed image data. The optimization module is used to construct the network structure and optimize it based on the desired results. The training module is used to process the preprocessed image data to obtain the source domain dataset for training. The optimized network structure is trained using the source domain dataset to obtain a model for accurate and automatic segmentation of knee joint images.

[0015] Thirdly, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a method for constructing a precise automatic segmentation model for knee joint images.

[0016] Fourthly, the present invention provides a storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a method for constructing a precise automatic segmentation model for knee joint images.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention acquires and preprocesses historical knee MRI image data to obtain preprocessed image data. This preprocessing ensures that images from different centers, devices, and scanning conditions are standardized before being used in training. This reduces the interference of differences in spatial scale, grayscale distribution, and image quality between multi-center heterogeneous data on model learning, improving the stability and repeatability of the training process and alleviating the performance degradation problem of existing methods when applied across centers. Simultaneously, the preprocessed image data increases the effective information density of the samples, allowing a limited number of samples to participate in training more stably even with a limited amount of high-quality labeled data, reducing learning bias and unstable performance caused by inconsistent sample quality. Furthermore, this invention constructs a network structure and optimizes it according to the desired results, making the network structure more suitable for the precise and automatic segmentation requirements of knee joint images. Under the same data conditions, it obtains more effective feature representation and more consistent segmentation output, reducing dependence on a single data distribution. This invention obtains a source domain dataset for training by processing preprocessed image data, and uses the source domain dataset to train the optimized network structure. This makes the organization of training data and the training process more standardized and controllable, and the model can learn robust segmentation-related features more easily. This improves the adaptability and generalization ability of the model to data from different sources, and solves the problems of insufficient adaptability to multi-center heterogeneous data and insufficient model reliability under the conditions of limited high-quality annotation scale. Attached Figure Description

[0018] Figure 1 This is a flowchart of the present invention; Figure 2 This is a system diagram of the present invention; Figure 3 This is a model construction diagram for Example 3; Figure 4 This is a technical roadmap for Example 4; Figure 5 This is a schematic diagram of the system in Example 5. Detailed Implementation

[0019] To further understand the content of this invention, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0020] Example 1: See Figure 1A method for constructing an accurate and automatic segmentation model for knee joint images includes the following steps: S1. Acquire historical knee MRI image data, preprocess the historical knee MRI image data, and obtain preprocessed image data.

[0021] S2, construct the network structure and optimize the network structure according to the desired results.

[0022] S3 processes the preprocessed image data to obtain the source domain dataset for training. The optimized network structure is trained using the source domain dataset to obtain a model for accurate and automatic segmentation of knee joint images.

[0023] Example 2: See Figure 2 A method for constructing an accurate and automatic segmentation model for knee joint images includes the following steps: The preprocessing module is used to acquire historical knee MRI image data, preprocess the historical knee MRI image data, and obtain preprocessed image data. The optimization module is used to construct the network structure and optimize it based on the desired results. The training module is used to process the preprocessed image data to obtain the source domain dataset for training. The optimized network structure is trained using the source domain dataset to obtain a model for accurate and automatic segmentation of knee joint images.

[0024] Example 3: See Figure 3 The specific method of this embodiment is as follows: Step 1: Acquire historical knee MRI image data, which includes samples of knee injuries of different types and degrees. Standardize the historical knee MRI image data to obtain preprocessed image data; the standardization process includes resampling the image spatial resolution, normalizing the image grayscale intensity, and performing noise suppression and artifact reduction.

[0025] Furthermore, the specific process of standardizing historical knee MRI image data is as follows: unifying the storage format of historical knee MRI image data and standardizing sequence naming and reading methods to complete format sequence organization; resampling the image data of different resolutions and slice thicknesses in the historical knee MRI image data with completed format sequence organization to a unified pixel spacing to complete resampling; normalizing the grayscale range of the resampled historical knee MRI image data and improving it to the required signal-to-noise ratio to obtain preprocessed image data.

[0026] The above methods unify the storage format, sequence naming and reading methods, pixel spacing, and grayscale intensity distribution of historical knee MRI image data from different sources, resolutions, and slice thicknesses. Simultaneously, noise suppression and artifact reduction improve the signal-to-noise ratio to the required level, thereby reducing the interference of multi-source data differences on subsequent model training and improving the consistency and usability of training samples. Specifically, unifying the storage format and sequence naming and reading methods improves the standardization of data management and batch reading, avoiding data mismatches caused by format and sequence confusion. Resampling image data of different resolutions and slice thicknesses to a unified pixel spacing reduces feature shifts caused by scale differences. Grayscale range normalization and improvement to the required signal-to-noise ratio enhances the learnability of tissue boundaries and texture features, thus contributing to improved stability of the segmentation model.

[0027] Step 2: Pre-construct Attention U-Net as the network structure, introduce an attention mechanism into the network structure to complete the construction of the network structure; integrate a multi-task learning strategy into the network structure so that the network structure can learn auxiliary feature recognition tasks related to organization recognition while performing organization segmentation, thereby completing the network structure optimization.

[0028] Furthermore, the optimized network structure includes an Attention U-Net backbone network. The encoder of the Attention U-Net backbone network is used to extract multi-scale features from local texture to high-level semantics layer by layer, and the decoder is used to restore spatial resolution layer by layer and output segmentation results. Skip connections are used to fuse the detailed information of the encoder and the semantic information of the decoder. An attention mechanism is introduced at the feature fusion position of the encoder and decoder. The attention mechanism is used to perform attention weighting on the features from the encoder and suppress noisy features. The Attention U-Net backbone network is equipped with a multi-task learning strategy for segmentation task branches and auxiliary feature recognition task branches.

[0029] Compared to using only the basic U-Net structure, the above method introduces an attention mechanism that can weight features from the encoder and suppress noise features during the feature fusion process between the encoder and decoder. This improves the response to the target tissue region and reduces background interference, thereby enhancing the ability to identify and segment complex tissue structures. At the same time, by integrating a multi-task learning strategy and setting up a segmentation task branch and an auxiliary feature recognition task branch, the network structure can learn auxiliary feature recognition tasks related to tissue recognition while performing tissue segmentation. It can obtain more discriminative deep features by sharing feature representations, thereby improving the stability and consistency of tissue segmentation in cases of complex structures and blurred boundaries.

[0030] Step 3: Obtain the preprocessed image data and crop the effective regions within the preprocessed image data to obtain target data blocks containing knee joint tissue, which serve as the source domain dataset for training. By cropping the effective regions to obtain target data blocks containing knee joint tissue, the interference of irrelevant background regions on model training can be reduced, and the training computation can be lowered, allowing the model to focus more on the effective information of the knee joint tissue, thereby improving training efficiency and segmentation performance. The source domain dataset used for training is divided into K training sets and K test sets using the K training sets; the optimized network structure is trained using the K training sets; the trained network structure is validated using the K test sets to obtain K optimal models; the key indicators required for the K optimal models are calculated to obtain a model for accurate and automatic segmentation of knee joint images. By employing the K-fold cross-validation method to divide the source domain dataset into K training sets and K test sets, and then validating the trained network structure based on the K test sets to obtain K optimal models, and then calculating the key indicators of the K optimal models, the influence of randomness caused by a single data partition can be reduced, making the evaluation results more robust, thereby improving the reliability of the obtained model for accurate automatic segmentation of knee joint images.

[0031] Furthermore, when training the optimized network structure using K training sets, different historical knee MRI image data are used as training sets to cover different types and degrees of knee joint injury samples and data distributions from different sources during the training process. This reduces the model's dependence on a single data distribution and improves the model's robustness to different datasets and its generalization ability to different datasets.

[0032] Example 4: See 4. In this embodiment, a multicenter knee MRI image dataset is established, and the training group and validation group are constructed under the unified diagnostic evaluation criteria and the center-level grouping strategy. At the same time, the safety demonstration and risk contingency plan for MRI examination-related risks are carried out, and the research is successfully implemented by relying on the existing case scale and multimodal image analysis technology reserves.

[0033] Patient MRI data containing keywords such as meniscus injury, ligament injury, medial / lateral / meniscus injury, anterior / posterior cruciate ligament injury, and medial / lateral collateral ligament injury were retrieved from the PACS system of our hospital and other invited centers as the source of data. Cases with poor image quality and incomplete scan sequences were excluded. Knee MRI images that meet the standards were exported and archived to form a retrospective data pool. Patients suspected of knee joint injury in orthopedic outpatient clinics or inpatients were included in the prospective data pool. Patients with contraindications to MRI were excluded. Eligible patients were informed of the examination process and risks and obtained informed consent before being included in the prospective data pool to continuously expand the sample size and enhance the data distribution coverage.

[0034] The patient group and control group were set up. The knee joint diagnosis was independently evaluated by two physicians using a double-blind method according to the diagnostic criteria. When the two physicians disagreed, a third physician reviewed the evaluation again and the consensus was used as the final evaluation result. The control group selected MRI images of the unaffected knee joint of the patient or MRI images of healthy subjects to ensure the comparability of the injured and non-injured samples and improve the discriminativeness of the subsequent model learning. The data of the training group and the test group were grouped at the center level in a 1:1 ratio to simulate the real clinical cross-center application scenario and to evaluate the robustness and generalization ability of the model, and to reduce the performance overestimation caused by the bias of data distribution from a single center.

[0035] MRI is a non-invasive imaging examination with no ionizing radiation and high soft tissue resolution. It is of great value in the diagnosis of knee joint injuries, especially meniscus and ACL injuries, and has high sensitivity, specificity and accuracy, with a well-established clinical application. The main risk comes from contraindications to MRI. Strict screening is required before the scan, and patients must be informed of precautions during the examination and sign an informed consent form. During the scan, the situation in the examination room must be closely monitored to ensure safety. For special populations such as women, children, and the elderly, as well as other special groups, the screening of contraindications, risk disclosure, monitoring and recording procedures should be strengthened to minimize the risks.

[0036] A retrospective collection of data from 4,323 patients with meniscus injuries and 2,156 patients with cruciate ligament injuries has been completed, and preliminary distribution statistics have been finalized. Additionally, data from approximately 3,600 cases have been collected from three other centers, and more centers will be continuously added to build a large-scale disease database. Regarding technical reserves, we have prior experience in researching the diagnosis and grading of brain diseases based on artificial intelligence and deep learning. Previously, we retrospectively collected multimodal MR data (T1c, T1, T2, multi-b-value DWI, and 3D-ASL) from 153 glioma patients, generating parametric maps and extracting 420 texture features and 90 histogram features. An SVM classifier for glioma grading prediction was established, validating that multimodal methods are superior to single-modal methods, and feature selection can reduce redundancy and optimize performance. The aforementioned experience in multimodal integration and modeling evaluation can be directly transferred to the organization of knee MRI data, model training evaluation, and multi-center generalization validation in this embodiment, thereby improving the feasibility and reliability of the research.

[0037] Example 5: See Figure 5 The present invention also provides an electronic device 100 for constructing a method for accurate automatic segmentation model of knee joint images; the electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on the at least one processor 102, and at least one communication bus 104.

[0038] The memory 101 can be used to store the computer program 103. The processor 102 implements the steps of the method for constructing a precise automatic segmentation model of knee joint images as described in Embodiment 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101. The memory 101 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0039] The at least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 102 may be a microprocessor or any conventional processor. The processor 102 is the control center of the electronic device 100, connecting various parts of the electronic device 100 via various interfaces and lines.

[0040] The memory 101 in the electronic device 100 stores multiple instructions to implement a method for constructing a precise automatic segmentation model for knee joint images, and the processor 102 can execute the multiple instructions to achieve the following: Historical knee MRI image data was acquired, and the historical knee MRI image data was preprocessed to obtain preprocessed image data. Build the network structure and optimize it based on the desired results; The preprocessed image data is processed to obtain the source domain dataset for training. The optimized network structure is trained using the source domain dataset to obtain a model for accurate and automatic segmentation of knee joint images.

[0041] Example 6: If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0042] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0043] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations 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, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0044] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0045] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for constructing a precise automatic segmentation model for knee joint images, characterized in that, Includes the following steps: Historical knee MRI image data was acquired, and the historical knee MRI image data was preprocessed to obtain preprocessed image data. Build the network structure and optimize it based on the desired results; The preprocessed image data is processed to obtain the source domain dataset for training. The optimized network structure is trained using the source domain dataset to obtain a model for accurate and automatic segmentation of knee joint images.

2. The method for constructing a precise automatic segmentation model for knee joint images according to claim 1, characterized in that, The specific method for acquiring historical knee MRI image data and preprocessing it to obtain preprocessed image data is as follows: Acquire historical knee MRI image data, which includes samples of knee injuries of different types and degrees; Historical knee MRI image data were standardized to obtain preprocessed image data. The standardization process included resampling the spatial resolution of the images, normalizing the grayscale intensity of the images, and suppressing noise and reducing artifacts.

3. The method for constructing a precise automatic segmentation model for knee joint images according to claim 2, characterized in that, The specific method for standardizing historical knee MRI image data to obtain preprocessed image data is as follows: The storage format of historical knee MRI image data was standardized, and the sequence naming and reading methods were standardized to complete the format sequence organization; Resampling was performed on historical knee MRI images with different resolutions and slice thicknesses that had been formatted and processed to a uniform pixel spacing. The grayscale range of the historical knee MRI image data that has been resampled is normalized and upscaled to the required signal-to-noise ratio to obtain the preprocessed image data.

4. The method for constructing a precise automatic segmentation model for knee joint images according to claim 1, characterized in that, The specific methods for constructing the network structure and optimizing it based on the desired results are as follows: A pre-built Attention U-Net is used as the network structure, and an attention mechanism is introduced into the network structure to complete the construction of the network structure; By integrating a multi-task learning strategy into the network structure, the network structure can learn auxiliary feature recognition tasks related to organization identification while performing organization segmentation, thereby completing network structure optimization.

5. The method for constructing a precise automatic segmentation model for knee joint images according to claim 4, characterized in that, The optimized network structure includes the Attention U-Net backbone network. The decoder of the Attention U-Net backbone network is used to extract multi-scale features from local texture to high-level semantics layer by layer. The decoder is used to restore spatial resolution layer by layer and output segmentation results. Skip connections are used to fuse the detailed information of the encoder and the semantic information of the decoder. An attention mechanism is introduced at the feature fusion location of the encoder and decoder. The attention mechanism is used to weight the features from the encoder and suppress noisy features. The Attention U-Net backbone network employs a multi-task learning strategy, used for segmentation task branches and auxiliary feature recognition task branches.

6. The method for constructing a precise automatic segmentation model for knee joint images according to claim 1, characterized in that, The preprocessed image data is processed to obtain the source domain dataset for training. The optimized network structure is then trained using the source domain dataset. The specific method for obtaining the accurate automatic segmentation model for knee joint images is as follows: The preprocessed image data is obtained, and the effective regions in the preprocessed image data are cropped to obtain the target data block containing the knee joint tissue, which is used as the source domain dataset for training. The source domain dataset used for training is divided into K training sets and K test sets using the K-fold cross-validation method. The optimized network structure is trained using K training sets; The trained network structure is validated using K test sets to obtain K optimal models; The key indicators required for the K optimal models are calculated to obtain a model for accurate and automatic segmentation of knee joint images.

7. The method for constructing a precise automatic segmentation model for knee joint images according to claim 6, characterized in that, When training the optimized network structure using K training sets, different historical knee joint MRI image data are used as training sets.

8. A method for constructing a precise automatic segmentation model for knee joint images, characterized in that, include: The preprocessing module is used to acquire historical knee MRI image data, preprocess the historical knee MRI image data, and obtain preprocessed image data. The optimization module is used to construct the network structure and optimize it based on the desired results. The training module is used to process the preprocessed image data to obtain the source domain dataset for training. The optimized network structure is trained using the source domain dataset to obtain a model for accurate and automatic segmentation of knee joint images.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for constructing a precise automatic segmentation model for knee joint images according to any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for constructing a precise automatic segmentation model for knee joint images according to any one of claims 1 to 7.

Images