Automatic evaluation of osteoarthritic cartilage severity
Figure 1: Overview of the automatic framework. Raw MRI inputs are segmented using a 3D-UNet to generate Cartilage Thickness Maps (CTh-Maps), which are then processed by a ResNet to output a severity score (CTh-Score).
Key Findings
- High Reproducibility: Both the Maps and the Score achieved an ICC > 0.98 in test/retest setups.
- Early Detection: The CTh-Score consistently detected cartilage degeneration progression up to 6 years before radiographic progression (KL grade) became apparent.
- Scalability: The framework is fully automatic and end-to-end, transforming raw MRI into standardized maps and quantitative scores.
- Trajectories: Analysis reveals three phases of progression: slow initial progression, accelerated degradation, and deceleration in end-stage disease.
Figure 2: CTh-Score trajectories illustrating the sensitivity to OA structural progression over time.
Related Publications
Additional work and resources related to CTh-Score and CTh-Maps: