Detection of evoked resonant neural activity in Parkinson's disease

Objective. This study investigated a machine-learning approach to detect the presence of evoked resonant neural activity (ERNA) recorded during deep brain stimulation (DBS) of the subthalamic nucleus (STN) in people with Parkinson's disease. Approach. Seven binary classifiers were trained to distinguish ERNA from the background neural activity using eight different time-domain signal features. Main results. Nested cross-validation revealed a strong classification performance of 99.1% accuracy, with 99.6% specificity and 98.7% sensitivity to detect ERNA. Using a semi-simulated ERNA dataset, the results show that a signal-to-noise ratio of 15 dB is required to maintain a 90% classifier sensitivity. ERNA detection is feasible with an appropriate combination of signal processing, feature extraction and classifier. Future work should consider reducing the computational complexity for use in real-time applications. Significance. The presence of ERNA can be used to indicate the location of a DBS electrode array during implantation surgery. The confidence score of the detector could be useful for assisting clinicians to adjust the position of the DBS electrode array inside/outside the STN.