Juggler's ASR: Unpacking the principles of artifact subspace reconstruction for revision toward extreme MoBI

To improve the Artifact Subspace Reconstruction (ASR) algorithm's performance for real-world EEG data by addressing the problem of low-quality or no calibration data identification in the original ASR (ASR_original) algorithm.

We proposed a new method for defining high-quality calibration data using point-by-point amplitude evaluation to eliminate collateral rejection of clean data, which is identified as the major cause of the problem with ASR_original. We compared non-parametric and parametric approaches, namely Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and the Generalized Extreme Value (GEV) distribution (ASR_DBSCAN and ASR_GEV, respectively).

We demonstrated the effectiveness of these approaches on simulated and real EEG data. Simulation results showed that ASR_DBSCAN and ASR_GEV removed simulated artifacts completely where ASR_original failed, both in time- and frequency-domain evaluations. In empirical data from 205-channel EEG recordings during a three-ball juggling task (n = 13), ASR_DBSCAN found 42 % and ASR_GEV found 24 % of data usable for calibration on average, compared to only 9 % by ASR_original. Subsequent Independent Component Analysis (ICA) showed that data preprocessed with ASR_DBSCAN and ASR_GEV produced brain ICs that accounted for more variance of the original data (30 % and 29 %) compared to ASR_original (26 %).

The proposed ASR_DBSCAN and ASR_GEV methods handle motion-related artifacts better than the original ASR algorithm, enabling researchers to better extract brain activity during real-world motor tasks. These methods provide a practical advantage in processing EEG data from experiments involving high-intensity motor activities, advancing biomedical research capabilities.