Improving the accuracy of current sagittal alignment evaluation system centered around pelvic incidence: a new machine-learning based classification.

Abstract

The study's aim was to characterize the variations in spinopelvic alignment among an asymptomatic population and to establish a classification system for this alignment. Additionally, it sought to formulate predictive formulas for lumbar lordosis (LL) based on pelvic incidence (PI) to enhance the accuracy of spinal balance assessments. This cross-sectional study included 726 asymptomatic individuals. Sagittal parameters were assessed through radiographic evaluation. Participants were categorized into clusters using K-means clustering. A decision tree incorporating PI and sacral slope (SS) was utilized to define the classification criteria. Linear regression models were developed to predict LL and PT, integrating the newly established classification. The sample was evenly divided into three clusters with distinct PI and LL averages. Cluster-specific predictive formulas for LL and SS were generated, highlighting the importance of spinopelvic alignment in spinal balance. For instance, in one cluster, the formula for LL was LL = 0.68*PI + 24.82, indicating a moderate correlation. The research successfully identified different patterns of sagittal balance and developed cluster-specific predictive formulas for LL based on PI. The findings underscore the significance of recognizing the anteverted pelvic subgroup for improving the precision of LL and SS predictions, which is vital for spinal surgery planning and achieving optimal sagittal balance.