Post-operative electrode placement prediction in deep brain stimulation using support vector regression

Deep brain stimulation (DBS) is a neurosurgical procedure for treating neurodegenerative diseases and neurological disorders such as Parkinson’s disease (PD) and epilepsy. Image guidance is crucial for the accurate placement of DBS electrodes. However, current surgical planning systems based on preoperative MR and CT images of the brain cannot take into account the intra-operative brain-shift, resulting in suboptimal electrode placement undermining clinical outcomes. In this study, a support vector regression (SVR) model was constructed based on 114 patient-specific data of PD patients. Two target nuclei were manually delineated based on pre-operative MR and CT images. Spatial coordinates of the two nuclei were collected and compared to the post-surgical electrode position from CT images. Analysis of a total of 45 features showed that the pre-operative target coordinates are the parameters mainly influencing the model prediction for both nuclei. The mean absolute error (MAE) of the prediction of the electrodes on unseen patients was 0.76mm. This study demonstrates the potential of using SVR modelling to improve current DBS surgical planning procedure and preoperative risk-assessment.