Using Intraoperative Microelectrode Recordings to Predict Improvement in Bradykinesia for Parkinson's Disease Patients

Abstract

Predicting the effectiveness of deep brain stimulation (DBS) for Parkinson's disease (PD) remains challenging due to the variability of individual responses. Local field potentials (LFPs) derived from intraoperative microelectrode recordings (MERs) within the subthalamic nucleus (STN) during DBS surgery reflect the neural activity associated with motor function in patients with PD. These signals can be utilized to assess the potential improvement index in response to therapeutic interventions. This study investigates the use of machine learning to predict DBS treatment outcomes prior to therapy. We present the evaluation of regression machine learning model to predict the improvement index of each PD patient using MER data. MERs were analyzed to extract significant features using Pearson correlation. A random forest regression model was able to achieve the optimal mean squared error (MSE) of 2.739 and coefficient of determination (R²) of 0.700 in the right-side improvement prediction. These findings suggest that MERs analysis combined with machine learning demonstrates potential for initially predicting personalized prediction of DBS outcomes in PD patients.Clinical relevance - These findings may enable clinicians to better predict individual patient responses to DBS therapy, potentially leading to more personalized treatment plans and improved clinical outcomes for people with PD.