Data-driven modeling for scoliosis prediction

Abstract

Traditional medical therapies for scoliosis are mostly based on the experience and intuitions of doctors, which does not guarantee the effectiveness of the treatment. Scoliosis prediction is of great significance to reduce the uncertainty for doctors on deciding the optimum treatment for patients. The paper aims to develop a prediction model to help physicians to make right decisions for an appropriate treatment. The change of Cobb angle in a definite period, which reflects the progress of scoliosis, is commonly considered as indication of scoliosis severity. The present study proposed several prediction models of scoliosis progression based on time series analysis and general regression methods. Performances of different time series methods as well as different general regression models were compared by the root mean square error (RMSE), standard deviation (SD) and the mean absolute percentage error (MAPE) as well as the Pearson product-moment correlation coefficient (r). The results show that the exponential moving average method performs better than any of the chosen time series methods and the linear regression model has higher predictive capability than any of the general regression models being compared.