Uncertainty Quantification of the Effects of Segmentation Variability in ECGI.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Medical Imaging

Background:

• Electrocardiographic Imaging (ECGI) techniques have advanced, but quantifying associated uncertainties, especially from geometric variations like segmentation, remains a challenge.
• Segmentation variability is a significant, yet under-explored, source of uncertainty in ECGI pipelines.

Purpose of the Study:

• To quantify the impact of segmentation variability on Electrocardiographic Imaging (ECGI) solutions.
• To utilize statistical shape modeling and uncertainty quantification (UQ) to assess ECGI pipeline uncertainties.

Main Methods:

• Developed a statistical shape model from nine patient segmentations using Shapeworks.
• Integrated the shape model into an ECGI pipeline and employed polynomial chaos expansion (PCE) for uncertainty quantification (UQ).
• Calculated uncertainties in pericardial potentials and local activation times (LATs) using UncertainSCI.

Main Results:

• Uncertainty in pericardial potentials correlated with shape model variability, notably at the heart's base and right ventricular outflow tract.
• ECGI demonstrated lower sensitivity to segmentation uncertainty in the posterior heart regions.
• Local activation time (LAT) calculations exhibited significant variability (up to 126ms standard deviation) due to segmentation, primarily in low conduction velocity areas.

Conclusions:

• The developed pipeline effectively visualizes ECGI uncertainty stemming from segmentation variability.
• This approach can guide researchers in mitigating or reducing segmentation-related uncertainties in ECGI.
• Statistical shape modeling and UQ are extendable to other computational modeling pipelines beyond ECGI.