Lateral Deep Brain Stimulation Prediction using Quantitative Susceptibility Mapping (QSM) and Unsupervised Gaussian Mixture Models

Motivation: Deep brain stimulation is a treatment for severe Parkinson's disease patients. Often, two electrodes are simultaneously implanted. As symptoms present asymmetric progression, evidence suggests unilateral implants effectively treat some patients. Candidate selection for unilateral versus bilateral procedures is challenging as many unilateral recipients require a second surgery within two years. Goal(s): Radiomic features from presurgical quantitative susceptibility maps are investigated for candidate selection, emphasizing patients with unilateral symptoms. Approach: An unsupervised Gaussian mixture model is fit to separate unilateral and bilateral distributions. Results: The model discriminates between candidates with accuracy $$$0.82$$$ and feature maps show distinction between candidate types. Impact: Preoperative QSM as a potential predictive tool for deep brain stimulation candidate selection is introduced. Findings suggest that imaging features complement clinical assessment in surgical planning. Future studies should examine additional targets to validate this approach.