Aims: Non-invasive fetal electrocardiography (fECG) is a promising clinical tool, yet extracting and interpreting the fetal signal remains challenging. This study performed sensitivity analysis (SA) of a subject-specific maternal-fetal ventricular electrophysiology (EP) model to identify the pregnancy-related parameters that most strongly influence fECG morphology.
Methods: The anatomical model was reconstructed from maternal-fetal MRI at 38 gestational weeks. Ventricular activity was simulated for 5 seconds using the pseudo-bidomain formulation, coupled with the adult ToR-ORd-dynCl and a fetal-adjusted ten Tusscher-Panfilov ionic model. A heart rate of 60 bpm and 134 bpm for the adult and fetal hearts were considered, respectively. A fascicular-based model for the His-Purkinje system (HPS) was employed for both hearts. Abdominal signals were extracted from body-surface potentials recorded at four electrode locations and the fECG was extracted using independent component analysis, following maternal ECG estimation and cancellation. SA considered 29 inputs, including fetal myocardial conductivities, fetal ionic conductances, fetal HPS root locations, and the conductivity of the amniotic fluid, blood pools, maternal organs, and subcutaneous fat. Linear parameter screening was applied, varying each input independently (±50% or within physiological bounds) and quantifying sensitivities via linear regression. Normalised indices ranked parameter influence across all six outputs considered.
Results: Fetal myocardial and blood pool conductivities were the primary determinants of QRS amplitude and duration. Ionic conductances, particularly the inward rectifier potassium current, dominated the T-wave amplitude and QTc. Sodium channel conductance emerged as the strongest driver of overall morphology changes, before and after fECG extraction. Among fetal HPS parameters, only the apico-basal coordinates had non-negligible effects.
Conclusion: This study identified parameters with negligible influence on the fECG (18 out of 29), which can therefore be fixed. This enables reduced model complexity for more robust variance-based SA and model calibration.