An Autonomous AI Framework for Knee Osteoarthritis Diagnosis via Semi-Supervised Learning and Dual Knowledge...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Computer-Aided Diagnosis

Background:

• Accurate knee osteoarthritis (KOA) diagnosis relies on classification models for disease severity assessment.
• Traditional methods face challenges with large labeled datasets, particularly in resource-limited developing countries.
• Data scarcity and patient privacy concerns hinder the development of robust KOA diagnostic tools.

Purpose of the Study:

• To propose PADistillation, a semi-supervised dual-knowledge distillation framework for KOA diagnosis.
• To address data scarcity and patient privacy issues in telemedicine and remote diagnostics.
• To enhance AI model performance in resource-constrained environments.

Main Methods:

• Employs attention-guided distillation to enhance student model learning with limited labeled data.
• Utilizes a personalized pixel shuffling method for dynamic, region-specific patient privacy protection.
• Leverages autonomous AI for optimization and real-time decision-making in resource-limited settings.

Main Results:

• Achieved 88.19% accuracy, 86.28% precision, and 86.94% F1 score with limited labeled data.
• Demonstrated over 2% increase in accuracy compared to mainstream semi-supervised methods.
• Improved training efficiency by 30% with a minimal 1.2% performance loss from privacy mechanisms.

Conclusions:

• PADistillation effectively enhances KOA diagnostic classification with limited data.
• The framework successfully balances performance, data scarcity, and patient privacy.
• PADistillation supports the expansion of telemedicine and remote diagnostics in underserved regions.