Objective: Our aim is to develop computational tools that can assist clinical decision in Deep Brain Stimulation (DBS) for Parkinson’s disease (PD) using intraoperative microelectrode recordings (MER). Different approaches were defined: 1) Construction of a generalizable and automatic tool for spike sorting and analysis of human MER with extraction of features; 2) Development of an unbiased classification model to distinguish STN from non-STN MER; 3) Identification of MER features that distinguish sensorimotor subdivision of STN vs. limbic and associative.

Background: DBS is a common treatment for advanced PD patients. Intra-operative MER along pre-planned trajectories are often used for accurate identification of STN, a common target for DBS-PD. However, this identification can be difficult in regions of transition and misidentification can lead to suboptimal location of the lead and inadequate clinical outcomes.

Method: Tools for unsupervised analysis, spike-sorting and extraction of related features of human MER were developed. A machine-learning classification model for high-accuracy identification of STN was programmed, using MER time and frequency properties. Neurophysiological characteristics of segregated STN segments were compared by dividing the STN in dorsal and ventral portions, which present higher probability of being motor (area of major interest within the STN) and non-motor regions respectively. Ongoing work will refine the results using anatomical gold standard through lead trajectory reconstruction, fused with an STN functional subdivision atlas.

Results: Using leave-one-subject-out validation, classification accuracy for STN-DBS recordings is 96.3±3.15% (30 trajectories, 5 patients; compared to human expert classification). Significant preliminary differences between STN segments were found (357 STN neurons in 5 subjects), as higher burst and firing rate of dorsal STN neurons (median (interquartile range) of 1.8(1.5) vs 1.15(0.05) bursts/s, p=0.001 and 21.4(16.85) vs. 15.3(14.33) spikes/s, p=0.013 respectively).

Conclusion: We’ve developed tools for human MER analysis, that provided good results in STN classification and are fast and generalizable for other brain regions. In line with the literature, preliminary activity differences were found in segregated STN segments. Ongoing anatomical work can further validate its’ usefulness in optimizing electrode placement and research purposes.

« Back to 2019 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/development-of-an-unsupervised-analysis-pipeline-for-human-microelectrode-recordings/