Real-Time ECG Artifact Removal for Adaptive Deep Brain Stimulation: A Comparative Study.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Adaptive deep brain stimulation (aDBS) for Parkinson's disease (PD) requires precise detection of beta oscillatory activity.
• Electrocardiographic (ECG) artifacts commonly contaminate local field potentials (LFPs), hindering the performance of control algorithms.
• The real-time feasibility of existing offline ECG artifact removal methods for aDBS is not well-established.

Purpose of the Study:

• To evaluate the performance of four ECG removal algorithms for real-time aDBS applications.
• To assess artifact removal quality and real-time processing feasibility of different algorithms.

Main Methods:

• Four ECG removal algorithms were tested: template subtraction (TS), singular value decomposition (SVD), extended SVD (eSVD), and Perceive toolbox (PR).
• Simulated datasets with varying contamination levels (-30 to +20 dB) and clinical recordings from 20 PD patients were used.
• Performance metrics included beta power preservation, signal-to-noise ratio, and 99th percentile processing latency (P99 < 50 ms for real-time feasibility).

Main Results:

• Template subtraction (TS) and standard SVD met the real-time feasibility threshold (P99 < 50 ms).
• TS demonstrated superior timing consistency (P99 ≈ 10 ms), while eSVD and PR exceeded the latency threshold (P99 > 90 ms).
• TS maintained beta power accuracy within ±12% across contamination levels, offering the best balance between artifact suppression and signal preservation.

Conclusions:

• Template subtraction (TS) is the recommended algorithm for real-time ECG artifact removal in aDBS for Parkinson's disease.
• TS provides a critical balance of computational efficiency and biomarker fidelity necessary for closed-loop control systems.
• This method ensures reliable detection of beta oscillatory activity essential for effective aDBS therapy.