Detection Method of Athlete Joint Injury Based on Deep Learning Model

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate knee joint MRI segmentation is crucial for clinical practice, but current methods require substantial expert labeling.
• The high workload associated with manual annotation limits the scalability and efficiency of knee MRI analysis.

Purpose of the Study:

• To develop a semisupervised learning model for knee joint MRI segmentation that reduces reliance on large, high-quality labeled datasets.
• To improve the accuracy and efficiency of knee MRI segmentation, thereby decreasing clinical workload.

Main Methods:

• A novel semisupervised learning segmentation network based on 3D scSE-UNet was proposed.
• The model incorporates a self-training framework with a cSE-block+ module to enhance feature representation and preserve edge information.
• A fully connected conditional random field was integrated to refine pseudolabel edges during training.

Main Results:

• The proposed 3D scSE-UNet model achieved segmentation performance comparable to fully supervised methods.
• The model demonstrated effectiveness in reducing the dependence on expert-labeled data for knee joint MRI segmentation.
• Validation was performed using the open-source MRNet and OAI datasets.

Conclusions:

• The developed semisupervised learning approach significantly lowers the demand for expert-labeled knee MRI data.
• The 3D scSE-UNet model offers an efficient and accurate solution for knee joint MRI segmentation.
• This method has the potential to alleviate the workload of clinical professionals in medical image analysis.