An Investigative Study Exploring Machine Learning Approaches for Optimizing Deep Brain Stimulation Programming

Esther Ademola, Martin Reich, Magda Gregorova

Abstract: This investigative study explores machine learning models for predicting dystonia improvement scores in Deep Brain Stimulation (DBS). Leveraging data from 85 subjects across seven European DBS centers, we employ various linear and non-linear modeling approaches. In contrast to previous studies utilizing probabilistic mapping, our direct utilization of the actual dataset yields improved results. The random forest model emerges as the most accurate predictor, with a mean deviation of 9.54 ± 6.08%. This implies that for a patient with an improvement score of 78%, the model predicts an improvement between 68% and 87%. This advancement in predictive accuracy holds potential implications for refining DBS programming, ultimately enhancing therapeutic outcomes for individuals with dystonia. In addition, regularization techniques play a pivotal role in determining feature importance thereby contributing to a nuanced understanding of factors influencing DBS therapy outcomes.

External IDs:dblp:conf/mdis/AdemolaRG24