Can the heartbeat-evoked potential (HEP) be separated from Cardiac Artefact (CA) using beamforming?

The heartbeat-evoked potential (HEP), a cortical response to heartbeats and neural marker of interoception, is clinically relevant but heavily contaminated by scalp cardiac-artefact (CA), making reliable distinction challenging. Because HEP and cardiac potentials are anatomically distinct, they may be separable via beamforming, a source localisation method isolating brain activity while suppressing external interference. Here, the first known validation of EEG beamforming for HEP source reconstruction was attempted, quantifying waveform recovery and spatial accuracy using simulated data. Linearly constrained minimal variance (LCMV) beamforming was used (A) to test signal recovery, 128-channel EEG was simulated for three models with known HEP waveforms: (1) a single right insula (R-Ins) HEP; (2) temporally distinct HEPs in R-Ins and right anterior cingulate cortex (R-ACC); (3) overlapping HEPs in both regions. Recovery was assessed by correlating reconstructed and known waveforms. (B) To test CA suppression, CA from brain-dead isoelectric EEGs was integrated to simulations, with varying source (-10-50dB) and sensor (0-30dB) signal-to-noise ratios (SNR). (C) To analyse empirical EEG from hypertensive and anxiety individuals (n=106). Null-space projection, using subject-specific QRS waveforms and its temporal derivative, removed residual CA in source waveforms. In model-1, beamforming achieved near-perfect recovery without CA (r>0.99, error=0mm) at -30dB source SNR, remaining robust in the presence of CA (r=0.72-0.94, error=0-5.7mm) but degrading below -20dB (r<0.3, error=27mm). Models 2 and 3 showed similar recovery but introduced R-ACC-to-R-Ins leakage. In empirical data, beamforming revealed significant HEP activity in R-Ins and R-ACC (p<0.001). A significant late R-ACC HEP difference (250-500ms post-R-peak) between hypertension and controls emerged after excluding low-SNR subjects and strengthened following QRS cleaning (p=0.035, d=0.96). LCMV beamforming reliably recovers simulated HEPs while suppressing CA given sufficient SNR. Applied empirically, beamforming detected clinically meaningful group differences in interoceptive regions (R-Ins/R-ACC), offering a source-level approach to separate HEP from CA that complements sensor-level methods.