This study evaluates the impact of different electrode placement strategies on 12-lead ECG simulations using MRI-derived 3D torso models. We compare a fully automated pipeline based on a statistical shape model (SSM) and machine learning segmentation, with a semi-automatic method involving manual refinement by clinical experts using an interactive tool.
Both methods were applied to the same MRI dataset of a 71-year-old male patient with dilated cardiomyopathy. The heart mesh was kept consistent, and simulations were run using the monodomain model implemented in MonoAlg3D, combined with the Ten Tusscher cellular model. A single apical stimulus was applied, and synthetic ECGs were generated using a pseudo-ECG algorithm.
Spatial analysis revealed small deviations between manual and automatic placements in precordial leads (V1–V6), but large discrepancies were observed in limb leads, particularly the LL and RL electrodes, which differed by over 300 mm. These differences were reflected in the ECG signals. Pearson correlation and relative RMSE (rRMSE) metrics showed that while some leads (e.g., aVF) maintained high waveform similarity, others (e.g., V6) presented significant mismatches. Sensitivity analysis—where electrodes were replaced one at a time—demonstrated that limb electrodes affected multiple channels, whereas precordial electrodes mainly influenced their own leads.
This work highlights that even small positional variations in electrode placement can cause notable differences in ECG morphology, potentially impacting clinical interpretation. While the automated pipeline offers speed and reproducibility, it may introduce significant errors in limb electrode localization without skeletal landmarks such as the sternum and ribs. Future work will include CT imaging and Monte Carlo-based perturbation studies to better quantify placement uncertainty.