Early detection of Osteoporosis in knee X-ray images by using Deep Learning Techniques

Abstract

Osteoporosis, a condition that degrades bone quality and leads to fractures, is a major health concern. Early diagnosis is crucial but often difficult due to high diagnostic costs and the challenges of manually interpreting X-ray images. This research addresses these issues by using a deep learning-based VGG-19 model to classify knee X-rays into three categories: normal, osteopenia, and osteoporosis. The study aims to (i) present a medically validated dataset of knee X-rays, and (ii) propose a transfer learning approach to classify disease stages. The dataset includes three subsets: the first with 239 images (36 normal, 54 osteopenia, 49 osteoporosis), the second with 744 images (372 normal, 372 osteoporosis), and the third with 372 images (186 normal, 186 osteoporosis). The VGG-19 model achieved a high accuracy of 90% across these datasets. Precision for the "Healthy" class was 0.93, recall was 0.92, and the F1-score was 0.92. The model is reliable for detecting osteoporosis, with a slight optimization needed for the "Osteopenia" class. This approach can help clinicians detect osteoporosis early, reducing fracture risks and improving patients' quality of life by enabling timely treatment.