Automated Detection of Knee Osteoarthritis Using an Enhanced AlexNet with Attention and Grad-CAM Visualization

Abstract

This paper proposes the Enhanced-AlexNet-ROI-Attention (EARA) framework for automated knee osteoarthritis (OA) detection. EARA integrates region-of-interest localization, attention mechanisms, and class-weighted optimization to improve sensitivity to subtle OA features. Evaluated on <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\sim 3,000$</tex> knee radiographs from a multi-center dataset, our method achieved 87.4% accuracy, 0.84 macro <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$F 1$</tex>-score, 0.81 weighted Cohen's kappa, and 0.92 AUC for binary OA detection (KellgrenLawrence <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\geq 2$</tex>), significantly outperforming AlexNet, VGG-16, and ResNet-50 (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$p&lt;0.05$</tex>). Grad-CAM visualizations confirm the model focuses on clinically relevant regions like joint space narrowing. The results demonstrate EARA's potential as an interpretable, robust tool for automated OA assessment that could reduce diagnostic variability in clinical practice.

Ähnliche Arbeiten