Scoliosis Detection Based on Feature Extraction from Region-of-Interest

Abstract

In recent years, the incidence of scoliosis is rising among adolescents. Considering the radiation hazards of X-ray detection, this paper intends to develop an effective non-radiation detection method for scoliosis. The research method consists of the following steps: (1) Collect clear an image of the back of the patient with a high-resolution digital camera, and optimize the image through preprocessing; (2) Segment the region of interest (ROI) of the back and spine to reduce the complexity of subsequent calculations; (3) Extract the back contour and mark the feature points; (4) Extract features according to the grayscale change of the spine ROI, and fit the spine midline according to the feature points; (5) Evaluate the degree of scoliosis according to the symmetry of the posture features and the Cobb angle of the spine midline. Finally, experimental results were analyzed, which indicate that the proposed scoliosis detection method can preliminarily evaluate the posture features. The scoliosis detection error fell in the reasonable range (0-4 degrees), when the subjects had a Cobb angle between 0 and 30 degrees. Hence, our algorithm is accurate and effective, and provides a low-cost, efficient solution for scoliosis detection.