Machine Learning Based Osteoporosis Detection

Osteoporosis is a bone disease that leads to the weakening of bones and makes them susceptible to fracture, especially in the elderly. Dual-Energy X-ray Absorptiometry (DEXA) is the best technique to measure bone density, but it is expensive and hard to obtain, thus early diagnosis becomes challenging. In this paper, we propose a mechanism to automatically identify knee osteoporosis from X-ray images using deep learning. The technique involves the transfer learning model of the Xception architecture and custom CNN models for identifying osteoporosis, osteopenia, and normal bone density in two-class and multiclass. The data were augmented from four publicly available data sets and further augmented using methods like normalization, class balancing, and augmentation. Xception had the best accuracy of 90.8% in multiclass and high precision and recall. Comparison with other models, VGG-19, ResNet, and InceptionNet, validates the potency of this method. The outcome validates the potential of deep learning as an affordable but effective tool for osteoporosis screening, especially in developing nations.