biv-me: Open-source software for generating time-varying biventricular meshes from cine cardiovascular magnetic resonance imaging with multi-cohort validation.

Abstract

The generation of geometric representations of the heart is essential for personalised approaches to cardiac assessment. Structured biventricular meshes customised to imaging data have demonstrated utility in a number of model-based applications that can provide more sensitive insights into patient health than routine cardiac indices alone. Cardiovascular magnetic resonance (CMR) imaging is a common starting point for the creation of digital twin geometries, with numerous published methods for mesh reconstruction. However, the majority of these methods are not open-source, are typically developed and validated using data from a single-centre, and lack deployability across heterogeneous scanning protocols and patient groups. We present an open-source, end-to-end pipeline (biv-me), to automatically generate time-varying biventricular meshes from cine CMR DICOM images, and perform external validation against a clinical reference software tool on 1313 CMR imaging studies across five publicly available datasets. We report excellent agreement in left and right ventricular indices and high scan-rescan reproducibility. Mesh generation was rapid, with a mean processing time of 2.5 min, and highly feasible, with 99% of meshes successfully generated to a high standard with median error of <1.5 mm. The biv-me pipeline - including code, models, and documentation - is available at https://github.com/UOA-Heart-Mechanics-Research/biv-me.