SEPP&M: A Machine Learning Dataset for Spine Surgery Somatosensory Evoked Potentials Prediction

The purpose of this study is to build a structured machine learning (ML) dataset for somatosensory evoked potentials (SEP) perdition for spinal surgery monitoring. To this end, the study collect and investigate the clinical records of 600 surgeries, and a series of manipulations is performed. First, based on the observations of raw SEP data frame, records without ground truth, invalid features and features with many missing values are removed. Second, new features are then constructed from the difference between the two successive measurements. Finally, ML algorithm K-nearest neighbor (KNN) and feature selection method max-relevance and min-redundancy (MRMR) are used to refine the remaining features. The resulting structured dataset is named as SEP prediction and monitoring (SEPP&M) dataset. The experimental results show that the KNN model trained on SEPP&M has higher accuracy than the one trained on the raw SEP dataset.