Deep Learning-Based Approach for Early Detection of Osteoporosis Using X-ray Imaging

Abstract

The main cause of factures in women and older adults after menopause is osteoporosis, a disorder that weakens bones. The start of an inexpensive technology for simpler detection is required because of the current expense of diagnosis and treatment procedures. Although X-ray imaging is widely available and reasonably priced, it has drawbacks in terms of computationally demanding feature extraction and manual interpretation. A CNN-based method that used transfer learning was used to categorise knee X-ray pictures into three groups: normal, osteopenia, and osteoporosis. A dataset of 381 knee radiographs to T-score from Quantitative Ultrasound, which has been medically validated, was used to assess four pre-trained CNN architectures: AlexNet, VGG16, ResNet, and VGG-19. The AlexNet model outperformed non-pretrained models by a vast margin, attaining 91.1% accuracy, anerror ratio of 0.09, and a validation loss of 0.54.