WMCA-Net: Wavelet Multi-Scale Contextual Attention Network for Segmentation of the Intercondylar Notch

Area of Science:

• Medical imaging analysis
• Orthopedic surgery
• Biomedical engineering

Background:

• Accurate segmentation of the femur intercondylar notch is vital for diagnosing knee diseases, surgical planning, and anterior cruciate ligament (ACL) reconstruction.
• Challenges in MRI segmentation include anatomical heterogeneity, similar tissue interference, and blurred boundaries, which standard semantic segmentation methods fail to fully address.

Purpose of the Study:

• To develop an advanced deep learning model for precise intercondylar notch segmentation in MRI images.
• To overcome limitations of existing methods in handling high-order internal variations and low-contrast features.

Main Methods:

• Proposed a novel Wavelet Multi-scale Contextual Attention Network (WMCA-Net).
• Integrated Shallow High-frequency Feature Dense Extraction Block (SHFDEB) for detailed feature extraction and Wavelet Split and Fusion Block (WSFB) for multi-resolution feature processing.
• Employed Multi-scale Depth-wise Convolution Block (MDCB) for cross-scale feature capture and Contextual-Weighted Attention Module (CWAM) for semantic association in uncertain regions.

Main Results:

• WMCA-Net achieved a Dice Similarity Coefficient of 93.16% on the intercondylar notch dataset.
• The model demonstrated a 95% Hausdorff Distance of 1.42 mm, indicating high segmentation accuracy.
• The network effectively handled anatomical variations and pathological changes like osteophyte formation.

Conclusions:

• WMCA-Net exhibits superior performance in segmenting the femur intercondylar notch from MRI data.
• The proposed model offers advanced anatomical adaptability and precise localization, benefiting clinical applications in knee joint assessment and treatment planning.