Deep Learning-Based OSIS-NET Framework for Accurate Detection and Classification of Osteoporosis Using MRI Imaging and Advanced Machine Learning Techniques

Abstract

Osteoporosis manifests in people as a painful bone disease which causes weakness because of reduced bone density and breakdown. A predictive diagnosis of osteoporosis and fragility fractures becomes possible during the first years of menopause. Two different types of bone loss in women develop osteoporosis through both menopause-related sources and external factors. DEXA diagnostic tools fail to demonstrate the intricate bone structural changes that occur in tissues effectively. This document proposes the implementation of OSIS-NET which detects and classifies osteoporosis conditions using deep learning principles. A collection of MRI images originates from the OASIS-3 Datasets followed by noise reduction using adaptive gaussian trilateral filter for artifact removal. The Sobel edge detector receives the images which underwent denoising to produce specific and detailed edges in dual images. The next stage applies deep learning (DL) based Dilated leaky ShuffleNet to process images for extracting textural and shape features from them. The selected ST features undergo an evaluation process using ML models which includes SVM, KNN, DT, and RF for detecting normal and abnormal patterns. The system reaches an excellent accuracy level that exceeds 99.76% for normal cases and 99.89% for abnormal cases. The model achieves 0.86% superior accuracy when compared to GoogleNet alongside 0.54% and 0.16% and 1.07% superior results than ShuffleNet, AlexNet and MobileNet respectively.