Hybrid Fuzzy Harmony Search and Multi-Modal Feature Fusion for Knee Osteoporosis Diagnosis

Abstract

Osteoporosis is a common bone disease and a major contributing factor to fractures among the elderly, directly impacting the quality of life of those affected. Therefore, early diagnosis is crucial. We proposed developing an intelligent system for detecting and classifying knee osteoporosis by analyzing knee X-ray images. We proposed a novel feature extraction technique called Hybrid Multimodal Feature Fusion (HMMFF), which combines traditional techniques such as GLCM, HOG, and ORB with deep learning models, including Squeezenet, YOLOv11, VGG-16, and ResNet50. The proposed system was applied to two databases for binary and multi-class classification. To improve the classification efficiency of the proposed system, feature selection was performed using the harmony search optimization algorithm. We also proposed a new hybrid feature selection method, the Hybrid Fuzzy Harmony Search Algorithm (HFHSA), that combines harmony search with Gath-Geva fuzzy clustering. The two techniques were evaluated and compared by feeding the selected feature sets into several machine learning classifiers. The results showed the superiority of the HFHSA technique for feature selection over Harmony Search alone, as the HistGradientBoosting model achieved the highest accuracy across the binary classifiers, reaching 96.8% in binary classification while the XGBoost model achieved 95.38% in multi-class classification. This performance improvement demonstrates that applying the HMMFF feature extraction and fusion technique, along with feature selection using HFHSA, confirms the effectiveness of integrating different AI techniques by extracting and selecting important features that contribute to improving the accuracy of classifying and diagnosing osteoporosis using input X-ray images.