Impact of Microelectrode Recording Artefacts on Subthalamic Nucleus Functional Identification via Features-Based Machine Learning Classifiers

Microelectrode recording (MER) signals validate planned trajectories during deep brain stimulation (DBS) surgery, ensuring accurate positioning of electrodes within the brain structure of interest. MERs can be classified through automatic algorithms that support the surgical procedure enhancing its efficiency. As a matter of fact, extracting relevant features from the signal is fundamental for the correct MER classification. However, MERs are affected by several artefacts that reduce the signal-to-noise ratio potentially compromising the accuracy of automatic techniques. The present study evaluates the influence of MER artefacts on the ability of intra-operative features-based machine learning (ML) models to identify signals from sub-thalamic nucleus (STN). Specifically, we applied two automatic artefact detection techniques on MER signals of 21 patients with Parkinson's disease receiving bilateral DBS. We compared the performance of ML classifiers trained a) with unprocessed data and b) with data processed using artefact detection. Overall, our model's performance is consistent with the one reported in the literature, achieving an average F1 score of 0.812 across models. We observed that models trained on data processed with artefact detection techniques demonstrated superior performance compared to those trained on the unprocessed dataset.