Fig.1 shows 3 sets of 20-sec breath-held ECGs measured at positions (i), (ii) and (iii), utilizing an MRI-compatible Cardiolab-IT digital ECG-recording system [2]. The adaptive filtering procedure was tested in 5 healthy subjects, and 2 patients with Premature Ventricle Contractions (PVCs) and Atrial Fibrillation (AF). Validation was based on comparing the filter-derived ECG_real with ECGs measured periodically outside the MRI. The data processing block diagram (Fig. 2) includes training of adaptive Least-Mean-Square filters with ECG_real input (i), application of the trained filters to ECGs acquired in (ii) and (iii), which separates the V_MHD from ECG_real.

PVC patient’s results (Fig. 3): (a) unprocessed surface-lead V6, (b) extracted ECG_real, and (c) V_MHD. In (b) S-T segment voltage is restored, and the R-wave dominates for gating. Aortic-flow vortices (c) generate oscillating-polarity V_MHD, with V_MHD peaking during S-T segment. Cardiac beat-to-beat stroke volume (d) was estimated from time-integrated systolic V_MHD. PVC beats produce substantially lower stroke volume than during sinus-rhythm. AF patient results (Fig. 4): (c) Irregular V_MHD and (d) irregular stroke volume are due to ventricular-filling differences at varying heart rates (100-140bpm). Athlete subject results (Fig. 5): Filter tracking of rapid heart-rate changes from 44bpm to 87bpm is shown during a treadmill stress test performed inside the MRI. V_MHD (b) and stroke volume (c) increase with heart rate, suggesting that the cardiac output matches higher demand. A stroke-volume comparison of all subjects (Fig 6), derived from time-integrated systolic V_MHDs, demonstrates the measurement’s sensitivity to pathology.