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

Adaptive deep brain stimulation (aDBS) for Parkinson's disease (PD) relies on accurate detection of beta oscillatory activity. However, electrocardiographic (ECG) artifacts frequently contaminate local field potentials (LFPs), compromising control algorithms. While offline cleaning methods exist, their feasibility for real-time operation within the strict timing constraints of current sensing-enabled devices remains unknown.

We evaluated four ECG removal algorithms, template subtraction (TS), singular value decomposition (SVD), extended SVD (eSVD), and the Perceive toolbox (PR), on simulated datasets (contaminated at -30 to +20 dB) and clinical recordings from 20 PD patients. Algorithms were assessed for artifact removal quality (beta power preservation, signal-to-noise ratio) and real-time feasibility (99th percentile processing latency-P₉₉ < 50 ms).

Only TS and standard SVD met the real-time feasibility threshold, with TS achieving superior timing consistency (P₉₉ ≈ 10 ms). eSVD and PR proved incompatible with closed-loop requirements (P₉₉ > 90 ms). While eSVD yielded the highest artifact suppression at extreme contamination, it suffered from poor signal preservation at moderate levels. TS demonstrated the best balance, maintaining beta power accuracy within ±12% across clinically relevant contamination levels.

TS is the recommended method for real-time aDBS applications, offering a safety-critical balance of computational efficiency and biomarker fidelity.