CARACAS, a novel automated tool for Cardiac Artifact Removal in Absence of CArdiac Signal

EEG recordings can contain cardiac related artifacts. Independent Component Analysis (ICA) followed by removal of cardiac Independent Components (ICs) is a powerful and widely used strategy for artifact correction. Most existing methods for automatic labeling of cardiac ICs require simultaneously recorded ECG (e.g., to compute correlation with IC time course, such as CORR algorithm). However, ECG is not always available. To address this limitation, we developed CARACAS (Cardiac Artifact Removal in Absence of CArdiac Signal), a novel tool that identifies cardiac ICs using only the IC time courses. Because cardiac ICs exhibit temporal profiles highly similar to ECG signals, we applied one of our existing function (GitHub repository here) designed to detect cardiac events (PQRST) in ECG signals to each IC time course. Analysis of the detected events enabled the differentiation of cardiac ICs from non-cardiac ICs, where unrelated signal variations are incorrectly identified as cardiac events. Using the 375 EEG-ECG recordings of the open-source dataset OpenNeuro ds003690, we compared the performances of three algorithms: CARACAS, IClabel (a generic IC classifier which does not require ECG), and CORR algorithm. A total of 21,375 ICs were manually and automatically classified. CARACAS achieved high performance (sensitivity = 0.923, specificity = 0.991), substantially outperforming ICLabel (sensitivity = 0.135, specificity = 0.999) and approaching the performance of CORR (sensitivity = 0.970, specificity = 0.998). We present a reliable ECG-free algorithm for cardiac IC detection in EEG. CARACAS provides a practical solution when ECG is unavailable, and is implemented in SASICA toolbox (MATLAB).