Computer-aided King classification of scoliosis

Abstract

Surgical planning for scoliosis relies on the classification of spinal curve pattern. The widely used King classification is subject to human measurement and judgment errors. To reduce the variability of King classification, a computer-aided method is proposed and its reliability is evaluated in this study. At the first step, endplate inclination and position of each vertebra are measured by a computerized system on the posteroanterior radiograph. Based on these measurements, the apical and end vertebrae are identified and then the Cobb angles are calculated automatically. The system subsequently classifies the curve types automatically based on the computerized Cobb measurements, the vertebral inclination, and the analysis of the relationship between the center sacral vertical line, and the apical lumbar vertebra. Five observers participated in the experiments. The kappa statistic was used to evaluate the variability. Classifications of 105 scoliotic cases by the five observers showed that with the aid of our system, the average intraobserver and interobserver kappa values improved from 0.75 to 0.90 and from 0.66 to 0.86, respectively. Results indicate that the proposed computerized system can assist a surgeon with the King classification of scoliosis.