Automatic Detection and Classification of Scoliosis From Spine X-Rays Using Transfer Learning

Abstract

In the last few years, the data is increasing in healthcare day by day. A manual system cannot reach to handle this big amount of data. An AI approach is organized to detect different features of medical data with an accurate diagnosis of multiple diseases. This research work provides Automatic detection and classification of Scoliosis (ADCS) using X-ray images. The spine of a patient with scoliosis curves in an S- or C-pattern, which is a pattern in opposite directions. The spine bends while rotating and curls sideways, causing this three-dimensional abnormality. X-ray imaging is used to identify scoliosis; however, it has historically been difficult and timeconsuming for the lumbar, cervical, and thoracic spinal structures. The manual approach to calculating spine curvature gave poor results with the lowest degree of precision. The status of the spine's vertebrae must be accurately determined from medical images for several clinical uses of spinal x-rays imaging. It happens when noisy or irrelevant material occurs in spinal xrays images, it calculates curvature to be inappropriate. This research proposed an automated framework (ADCS) for detecting the curvature of the spine from the spinal column. According to the increase of a large amount of data in medical departments, deep learning models provide useful information from the images already gathered in primary care. Deep learning algorithms offer a quicker and more effective scoliosis identification method than manual X-ray analysis. Using a pre-trained EfficientNet (EN) model, scoliosis is detected and classified from X-ray images of spine curvature. In the first step, we attained an accuracy of 76% in the initial evaluation of the model without augmentation. In the second step, we used the augmentation technique with the same model and as a result, we were able to attain an accuracy of 89%. Our results show that spine curvature from spine x-rays may be detected and classified using the automatic scoliosis detection techniques that have been proposed.