Cartilage Auto-Segmentation of 3D T2* GRE Sequence in 7T High-Resolution 3D MRI

Motivation: Articular cartilage degeneration is the hallmark of knee osteoarthritis. Goal(s): Enable fast and robust assessment of degenerative changes in knee articular cartilage through auto-segmentation of high-resolution 7T 3D T2*-weighted MRI sequences. Approach: Train a supervised residual U-Net model on a limited ground truth dataset to generate preliminary segmentations, easing manual segmentation efforts and expanding the data pool efficiently. Results: The auto-segmentation achieved an overall mean Dice score of.826, 81.5 for the control cohort, 0.83 for patients with medial meniscus posterior root tears, and 0.817 for post-repair MRI of the same patients. Impact: This method accelerates cartilage segmentation in 3D T2*-weighted MRI, reducing manual correction, speeding ground truth creation, potentially supporting quantitative analysis, and enhancing efficiency in cartilage assessment for knee osteoarthritis.