An investigation of machine learning algorithms for prediction of lumbar disc herniation.

Abstract

The prevalence of lumbar disc herniation (LDH), which makes patients' daily activities more difficult and reduces their quality of life, has tended to increase recently. Many risk factors associated with LDH have been reported. In this study, LDH was predicted using machine learning techniques using measures of the lumbar paraspinal muscles, lumbar vessels cross-sectional area (CSA), and lumbar sagittal curve. Three hundred and forty-four individuals' MR scans were prospectively enrolled (264 with LDH and 80 healthy). Predictive factors were the lumbar sagittal curve and the cross-sectional areas of the lumbar paraspinal muscles and vessels from sagittal and axial MR images. The measurements have been analyzed via ten different and most common machine learning algorithms by considering a comprehensive parameter tuning and cross-validation process. The variable importance results have been also presented. XGBoost algorithm among all algorithms has provided the best results in terms of different classification metrics including f-score ([Formula: see text]), AUC ([Formula: see text]), accuracy ([Formula: see text]), and kappa ([Formula: see text]). The findings of this study demonstrated that cross-sectional areas of the quadratus lumborum and abdominal aorta can be utilized as a reliable indicator of LDH. Consequently, the developed model and the variables found to be important may guide to healthcare professionals to make more accurate and effective decisions in terms of prediction the LDH.