Semi-automated estimation of spinal curvature from scoliosis radiographs using difference matrix

Abstract

Scoliosis is a medical condition in which there is a deformity and bending of the spine. Early detection and corrective measures can prevent severe cases, mainly in early childhood. The gold standard method for the diagnosis of scoliosis is the measurement of spine curvature using Cobb angle. In this paper, a semi-automated approach for accurate estimation of Cobb angles from scoliosis radiographs using a difference matrix is proposed. Given the landmark points of the vertebral columns in the spine, the spinal mid-line is determined by curve fitting of the 4th degree polynomial. Using a difference matrix created by calculating slopes in key points along the curve, Cobb angles are calculated. Data used in this work is obtained from the Accurate Automated Spinal Curvature Estimation (AASCE) Challenge 2019. The method is evaluated using mean absolute error and symmetric mean absolute percentage error. The proposed method gives promising results and could be used in accurate estimation of Cobb angle for the diagnosis of scoliosis.