Classification of DBS microelectrode recordings using a residual neural network with attention in the temporal domain

Abstract

During the Deep Brain Stimulation (DBS) surgery for Parkinson's disease (PD), the main goal is to place the permanent stimulating electrode into an area of the brain that becomes pathologically hyperactive. This area, called Subthalamic Nucleus (STN), is small and located deep within the brain. Therefore, the main challenge is the precise localization of the STN region, considering various measurement errors and artifacts. In this paper, we have designed and developed a computer-aided decision support system for neurosurgical DBS surgery. The implementation of this system provides a novel method for calculating the expected position of the stimulating electrode based on the recordings of the electrical activity of brain tissue. The artificial neural network with attention is used to classify the microelectrode recordings and determine the final position of the stimulating electrode within the STN area. Experiments have verified the utility and efficiency of our system. The tests were carried out on many recordings collected during DBS surgeries, giving encouraging results. The experimental results demonstrate that deep learning methods extended with self-attention blocks compete with the other solutions. They provide significant robustness to recording artifacts and improve the accuracy of the stimulating electrode placement.

Keywords: Attention; Classification; Deep brain stimulation; Microelectrode recording; Residual neural network; Temporal domain.