ML-STIM: Machine Learning for SubThalamic nucleus Intraoperative Mapping.

Area of Science:

• Neurosurgery
• Computational Neuroscience
• Biomedical Engineering

Background:

• Deep Brain Stimulation (DBS) of the SubThalamic Nucleus (STN) effectively treats motor symptoms in Parkinson's Disease (PD).
• Intraoperative STN identification relies on MicroElectrode Recordings (MERs), a process that is operator-dependent, time-consuming, and prone to variability.
• Automating MER analysis is crucial for enhancing the efficiency and consistency of DBS procedures.

Purpose of the Study:

• To develop and validate ML-STIM, a machine learning pipeline for automated, real-time classification of MERs for intraoperative STN identification.
• To assess the accuracy and generalizability of ML-STIM across independent datasets.
• To demonstrate the computational efficiency and interpretability of the proposed ML pipeline.

Main Methods:

• Developed ML-STIM, a pipeline involving MER pre-processing, feature extraction, and MultiLayer Perceptron classification.
• Implemented an adaptive artifact removal algorithm to preserve STN signals while identifying artifacts.
• Selected MER features using correlation analysis and ReliefF ranking, then trained and validated on Dataset A (46 patients) and tested on Dataset B (36 patients).

Main Results:

• ML-STIM achieved high classification accuracy: 87.8 ± 1.7% on Dataset A and 83.8 ± 1.6% on Dataset B.
• The model significantly outperformed a state-of-the-art deep learning model (ResNet-AT, p < 0.01).
• Real-time processing of 10-second recordings was achieved in 139.4 ± 2.1 ms, and artifact removal significantly improved specificity (p < 0.001).

Conclusions:

• ML-STIM provides an accurate, interpretable, and computationally efficient solution for intraoperative STN identification.
• The pipeline demonstrates robust generalizability to data from different surgical centers.
• Automated MER analysis using ML-STIM has the potential to streamline DBS surgery for Parkinson's Disease patients.