Objective:

To introduce a semi-supervised learning framework for characterizing knee osteoarthritis (KOA) stages using combined MRI and CT-derived cartilage features, addressing the limitations of current subjective diagnostic methods.

Approach:

• Cohort Analysis: Analyzed a cohort of 133 knee scans, including 36 expert-labeled cases categorized as healthy, early degeneration, or advanced degeneration, to establish a baseline for the SSL framework.
• Semi-Supervised Learning: Utilized graph-based semi-supervised learning (SSL) methods, specifically Label Propagation and Label Spreading, to produce pseudo-labels for unlabeled samples, enhancing the training dataset.
• Performance Evaluation: Evaluated classifiers trained on the SSL-labeled dataset, achieving high weighted F1-scores with Support Vector Machines (0.84) and Logistic Regression (0.81), indicating robust performance.

Key Findings:

• Label stability across ten Monte Carlo runs showed high agreement (0.91 ± 0.14) and substantial reliability (Fleiss’ kappa = 0.781), indicating consistency in labeling.
• Supervised classifiers achieved robust performance with weighted F1-scores of 0.84 and 0.81 for Support Vector Machines and Logistic Regression, respectively, demonstrating the effectiveness of the SSL approach.
• Statistical analysis confirmed significant differences among the three KOA classes for all extracted features, using appropriate statistical tests to validate findings.

Interpretation:

The volume-to-surface ratio and density heterogeneity were identified as having the strongest discriminatory power, reflecting progressive cartilage thinning and structural changes associated with KOA, based on the results obtained.

Limitations:

• The study relied on a limited number of expert-labeled cases to seed the SSL framework, which may affect the robustness of the findings.
• The generalizability of the findings may be constrained by the specific cohort analyzed, limiting applicability to broader populations.

Conclusion:

Combining expert knowledge with SSL enables reliable KOA stratification, providing insights into cartilage degeneration and laying the groundwork for objective imaging-based biomarkers, as demonstrated by the study's findings.