Dual-Rater Noise Compensation in UPDRS Improvements from Deep Brain Stimulation via Quantitative Susceptibility Mapping Outcome Prediction

Motivation: Predicting deep brain stimulation surgery outcome is motivated by shortcomings of the levodopa challenge test. Goal(s): Preoperative quantitative susceptibility maps contain informative statistical information in the deep gray nuclei iron distributions. These distributions can be represented in the susceptibility radiomic features collected over the surgical targets such as the subthalamic nucleus. Approach: A noise compensation model is introduced to predict surgical outcomes in the presence of both rater noise and dual raters. Results: Use of this method improves the prediction based on presurgical imaging. Impact: The introduced noise compensated radiomic model allows for predictions on datasets with more than one rater, reflecting a clinically relevant setting in which candidate selection can be improved.