Osteoporosis Detection via Deep Learning

الملخص

Osteoporosis is a chronic skeletal disorder marked by reduced bone mineral density (BMD) and deterioration of bone microarchitecture, which significantly increases the risk of fractures. Timely and precise diagnosis is essential to initiate early treatment and prevent complications. Osteoporosis can be detected using several diagnostic methods, each varying in accuracy, accessibility, and clinical application. Deep learning can emerge as a tool for detecting osteoporosis by examining imaging data such as X-rays and DEXA scans for subtle indicators. It supports early diagnosis, benefits both patients and specialists, and enhances overall healthcare delivery. In this study, applicability of deep learning-based convolutional neural networks (CNNs)—specifically DenseNet201, ResNet50, and DenseNet121 — investigated for the automatic detection of osteoporosis from knee x-ray radiographic. The used CNN models were trained using transfer learning were adapted for binary classification (normal vs. osteoporotic). To enhance model generalization, standard preprocessing and different aspects of image augmentation techniques were applied. All used models are evaluated using common evaluation metrics and compared. Among the models used, DenseNet201 performed well across all performance metrics, achieving a classification accuracy of 85.11%, outperforming the other models. The study underscores the potential of CNNs to support radiologists in performing efficient osteoporosis screening. The results aid in diagnosing osteoporosis by revealing early signs of bone loss, allowing timely intervention. This supports treatment decisions and helps prevent fractures, ultimately improving patients' quality of life and enabling proactive condition management.