MV2SwimNet: A lightweight transformer-based hybrid model for knee meniscus tears detection

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Orthopedic Diagnostics

Background:

• Knee ailments, particularly meniscus injuries, affect millions globally, with over 14% of individuals over 40 experiencing related conditions.
• Conventional MRI interpretation is labor-intensive, prone to errors, and relies heavily on radiologist expertise, necessitating automated solutions.
• Existing deep learning methods using Convolutional Neural Networks (CNNs) struggle with long-range dependencies and global context in medical images.

Purpose of the Study:

• To develop an automated, highly accurate diagnostic tool for knee disease detection using medical imaging.
• To overcome the limitations of CNNs in capturing global contextual information in MRI scans.
• To introduce MV2SwimNet, a hybrid deep learning model designed for enhanced knee MRI analysis.

Main Methods:

• Proposed MV2SwimNet, a hybrid architecture integrating MobileNetV2 and Swin Transformer.
• Incorporated Window Multi-Head Self-Attention (W-MSA) for effective region attention in MRI scans.
• Utilized Multi-Stage Hierarchical Representation (MSHR) for progressive and robust feature learning across different representation levels.

Main Results:

• Achieved 99.94% accuracy on dataset1 and 96.04% accuracy on dataset2 using 3-fold cross-validation.
• Demonstrated superior performance compared to state-of-the-art techniques in knee MRI analysis.
• Validated the model's efficiency, robustness, and potential for real-world medical applications.

Conclusions:

• MV2SwimNet offers a highly accurate and automated solution for knee disease detection from MRI scans.
• The hybrid approach effectively integrates local and global features, enhancing diagnostic capabilities.
• The model shows significant potential for clinical application, improving the efficiency and accuracy of orthopedic diagnostics.