DBS Therapeutic Window Prediction

Machine learning framework for predicting therapeutic windows in Deep Brain Stimulation for Parkinson's Disease using magnetoencephalography and local field potentials.

Electrophysiological signatures predict the therapeutic window of deep brain stimulation electrode contacts

🧠 Overview

This repository implements a machine learning pipeline to predict the therapeutic window (TW) of electrode contacts in Deep Brain Stimulation (DBS) for Parkinson's Disease patients. By analyzing resting-state neural oscillations from the subthalamic nucleus (STN) and STN-cortex coherence patterns, the model helps identify optimal contacts for chronic stimulation.

Key Features

• Multimodal Analysis: Combines MEG and LFP recordings for comprehensive neural signatures
• Advanced Feature Engineering: Extracts power spectra and coherence features across frequency bands
• Prediction: XGBoost-based regression with nested cross-validation
• Automated Contact Ranking: Aim is to speed up monopolar review procedures

Architecture

DBS_Prediction_TW/
├── main.py                    # Main pipeline orchestrator
├── collector.py               # Data collection and TW calculation
├── collector_utils.py         # Helper for collector utilities
├── preprocessing.m            # MEG-LFP feature extraction (MATLAB)
├── fooof_lfp.m                # Apply FOOOF to LFP
├── predictor.py               # XGBoost model training & prediction
├── analyser.py                # Performance metrics & visualization
├── config.json                # Configuration parameters

Prerequisites
- Python 3.8+
- MATLAB R2019b+ (for preprocessing)
- FieldTrip toolbox

📄 Citation

If you use this code, please cite:

Rassoulou, F. et al. Electrophysiological signatures predict the therapeutic window of deep brain stimulation electrode contacts. npj Digit. Med. 8, 635 (2025).

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.