Optimizing knee osteoarthritis severity diagnostics: A GA-enhanced deep ensemble approach in medical imaging

Abstract

Knee osteoarthritis (OA) is a common degenerative condition that impairs mobility, particularly in older adults. Early and accurate diagnosis is vital for effective treatment and improved patient outcomes. This study proposes a deep learning model for automatic OA classification from X-ray images, optimized using a Genetic Algorithm (GA) to enhance performance. A diverse, expert-annotated dataset was used, with YOLOv8 employed for image cropping to focus on key knee regions. Multiple deep learning models, including SE-ResNeXt, ConvNeXt, and EfficientNet, were combined into an ensemble optimized by GA. The model achieved 95% accuracy in classifying OA severity levels (normal, mild, severe) and outperformed traditional diagnostic methods in accuracy and consistency. Grad-CAM visualizations highlighted critical diagnostic regions, supporting clinical interpretability. The proposed approach shows promise for assisting radiologists in efficient OA diagnosis, reducing workloads, and improving diagnostic precision. Further validation on larger datasets will ensure broader applicability.