Detection of Osteoporosis with DEXA Scan Images using Deep Learning Models
P. Varalakshmi, Smruthi Sathyamoorthy, V Darshan, V. Ramanujan, Sakthi Jaya Sundar Rajasekar · 2022 International Conference on Advances in Computing, Communication and Applied Informatics (ACCAI) · 2022
Osteoporosis is a chronic degenerative disease. Postmenopausal women and people over 50 years of age are at the highest risk of developing osteoporosis. If left untreated, osteoporosis can prove to be debilitating and might require hospitalization. So it is crucial to identify the risk of osteoporosis and take appropriate measures before the disease can negatively impact the quality of human life. Many existing osteoporosis classification models are computationally expensive and suffer from low accuracy due to inefficient processing of scan images. In this paper, the best model for predicting the risk of osteoporosis using DEXA scan images is identified. To improve the accuracy and specificity of the diagnosis, the scan images are pre-processed using denoising, image enhancement, and thresholding. Different CNN and hybrid models are trained, tested and the performance of the models is charted to infer the best model that gives promising results. Experimental results show that the proposed approach of pre-processing the images, data augmentation, sampling with Synthetic Minority Over Sampling Technique (SMOTE) and Inception v3 CNN model achieves the highest accuracy of 92.05%, specificity of 90.37%, the sensitivity of 93.87%, Precision of 89.97%, and F1 score of 91.88%.