Detection and Segmentation of Osteoporosis in Human Body using Recurrent Neural Network

Abstract

Osteoporosis is a disease that weakens the bone strength causing fractures and life-threatening complications. It has been estimated to affect more than 200 million people worldwide. We used recurrent neural network for the detection and segmentation of osteoporosis in human bones specially spinal cord. This technique produces an image representing bone density at the scanned site. However, bone density itself is only a moderate predictor of fracture risk, which creates demand for alternative prediction models. Deep learning, and especially recurrent neural networks, has been the leading image analysis approach in recent years. It has produced good results also in medical image analysis, including some orthopedically problems. This study seeks to discover if recurrent neural networks can predict osteoporotic fractures from spine images. By experimenting with different network architectures, the study aims to gain an understanding of the most promising design directions of such prediction models. In this context, also some practical deep learning challenges such as low training speed and lack of transparency are addressed.