A new approach for automated detection of behavioral task onset for patients with Parkinson's disease using subthalamic nucleus local field potentials

摘要

We present a new automated onset detection approach for behavioral tasks of patients with Parkinson's disease (PD) using Local Field Potential (LFP) signals collected during Deep Brain Stimulation (DBS) implantation surgeries. Using time-frequency signal processing methods, features are extracted and clustered in the feature space. A supervised Discrete Hidden Markov Models (DHMM) is employed and merged with Support Vector Machines (SVM) in a two-layer classifier to boost up the detection rate. According to our experimental results, the proposed approach can detect the onset of behaviors using LFP signals collected during DBS surgery with the accuracy of 84% while the acceptable delay is set to 1500 ms.