CMC-Net: 3D calf muscle compartment segmentation with sparse annotation.

Area of Science:

• Medical Imaging
• Deep Learning
• Musculoskeletal System

Background:

• Accurate 3D segmentation of calf muscle compartments in MR images is crucial for diagnosing and monitoring muscular diseases.
• Current deep learning methods require extensive annotated data, which is difficult and time-consuming to obtain.

Purpose of the Study:

• To develop a novel deep learning framework (CMC-Net) for efficient 3D calf muscle compartment segmentation.
• To minimize the requirement for expert-annotated data in medical image segmentation tasks.

Main Methods:

• CMC-Net utilizes an unsupervised approach to select the most representative 2D slices for expert annotation.
• Ensemble model training incorporates both annotated and unannotated slices for improved generalization.
• A pseudo-labeling strategy refines the model for accurate 3D segmentations.

Main Results:

• The proposed method achieves high segmentation performance with significantly reduced annotation ratios.
• CMC-Net outperforms existing state-of-the-art methods when full annotation data is utilized.
• Validation on a 3D MR thigh dataset demonstrates its effectiveness for segmenting various leg muscle groups with sparse annotations.

Conclusions:

• CMC-Net offers an effective solution for accurate 3D muscle segmentation in medical imaging, addressing the challenge of limited annotated data.
• The framework demonstrates the potential for more efficient and accessible diagnostic tools for muscular diseases.