Greylag Goose Optimization-Driven Convolutional Neural Network for Multi-Class Knee Osteoporosis Classification from X-Ray Images

Abstract

Knee osteoarthritis is one critical musculoskeletal state. Early detection of knee OA is a significant challenge in clinical settings. Automated analysis of knee X-ray images can provide a way forward for accurate OA diagnostic support. This study proposes a custom Convolutional Neural Network (CNN) optimized by the Greylag Goose Optimization (GGO) algorithm. The aim is multi-class classification of knee X-rays into Normal, Osteopenia, and Osteoporosis categories. The baseline CNN achieved an accuracy of 74.80%, sensitivity of 83.20%, specificity of 85.60%, precision of 74.10%, and F1-score of 73.60%. After the graylag goose optimization accuracy improved to 83.52% with gains across all metrics. The results indicate that nature based optimization is an effective strategy for accurate medical image diagnosis through deep learning methods for classification tasks