Osteo-Net: A Robust Deep Learning-Based Diagnosis of Osteoporosis Using X-ray images

Osteoporosis results in the deterioration of bone tissues and this problem is prevalent among people all over the world, especially older people. The diagnosis of osteoporosis is frequently made clinically manifested by fractures linked with bone fragility. Thus, early diagnosis would be required to take proper treatment and eliminate excess fractures, lowering mortality and morbidity. The development of deep learning-based technique for diagnosing osteoporosis disease from bone X-ray images, a commonly available and low-cost image-based medical examination approach, is the objective of this study. A deep learning-based architecture, Osteo-Net with many blocks and skip connections is presented in this work. The proposed technique utilizes the robustness of deep learning models to extract high-level features from low-quality X-ray images. The trained model achieved a validation accuracy of 84.06% and testing accuracy of 82.61% on unseen test images with less training time. The proposed method is low-cost and computationally efficient. The experimental results show an excellent classification performance when used for osteoporosis screening and the high efficacy of the proposed method over other state-of-the-art methods. The proposed low-cost deep neural network-based approach could be utilized as a supplement to Dual-energy X-ray Absorptiometry (DXA) screening, particularly in primary health care centers with insufficient DXA machines.