Automatic Detection and classification of Scoliosis from Spine X-rays using Transfer Learning

Abstract

Scoliosis is a typical spinal disease that causes the spine to curve. Early treatment during the formation of the spine can greatly reduce the chances of health issues. Diagnosis of scoliosis relies on X-ray imaging, using X-ray images to diagnose lumbar, cervical, and thoracic spinal structures have traditionally proven difficult and time-consuming. Many clinical applications of spinal imaging require the accurate and robust identification of vertebrae from medical images. This paper presents an automated approach using deep learning to detect the spine’s curvature using its spinal column. Models of deep learning could be used to assist with the increasing volume of medical imaging data as well as provide initial interpretation of images gathered in primary care. Deep learning algorithms are a quicker and more efficient alternative to manual X-ray investigation for scoliosis detection. X-ray images of spine curvature are used to detect and classify scoliosis using a pre-trained EfficientNet model. In the first stage, the model was evaluated without augmentation, in which we achieved an accuracy of 78 %. In the second step, we augment the training data by using machine learning techniques, and after that, we achieved an accuracy of 86 %. Our findings show that the proposed automatic scoliosis identification method can accurately detect and classify spine curvature in X-ray images.