An AI system for continuous knee osteoarthritis severity grading: An anomaly detection inspired approach with few labels

2024-07-16

Artificial Intelligence in Medicine

AI Summary

Current grading systems for knee osteoarthritis (OA) suffer from subjectivity and limited accuracy, while mainstream automated approaches rely heavily on large-scale annotated datasets and fully supervised training, perpetuating existing assessment biases. To address these limitations, we propose the first weakly supervised, end-to-end framework for continuous OA severity assessment. Our method innovatively reformulates OA grading as an anomaly detection task, decoupling pathological representation learning from ordinal severity regression. It integrates self-supervised contrastive learning, manifold-constrained variational reconstruction, and density-guided ordinal regression—enabling unsupervised pretraining followed by fine-tuning with minimal labeled data. Evaluated on multi-center MRI datasets, our framework achieves a Kendall Tau of 0.89, reduces annotation requirements by 90%, and significantly outperforms fully supervised baselines.

Technology Category

Data Mining & Knowledge Management: Anomaly/Outlier DetectionMachine Learning: Unsupervised & Self-Supervised LearningKnowledge Representation and Reasoning: Ontologies

Application Category

Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingEconomics, Online Markets and Human Computation: Data quality aspects of human-annotated datasetsGraph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Problem

• Automated continuous grading of knee osteoarthritis severity
• Reducing reliance on large annotated datasets for training
• Improving accuracy over existing ordinal grading systems

Innovation

• Self-supervised anomaly detection with limited data
• Pseudo labeling and denoising using CLIP
• Dual Centre Representation Learning for grading