Impact of Contour Boundary Offsets on 4D Flow CMR-Derived Intracardiac Haemodynamic Parameters.
Authors
Affiliations (5)
Affiliations (5)
- Norwich Medical School, University of East Anglia, Norwich NR4 7TJ, UK.
- Cardiology Department, Norfolk and Norwich University Teaching Hospitals, Norwich NR4 7UY, UK.
- Department of Molecular Medicine and Surgery, Karolinska Institute, SE-171 76 Solna, Sweden.
- Leeds Institute of Cardiovascular and Metabolic Medicine, University of Leeds, Leeds LS2 9JT, UK.
- Leiden University Medical Centre, 2333 Leiden, The Netherlands.
Abstract
Four-dimensional (4D) flow cardiovascular magnetic resonance assesses advanced haemodynamic parameters like kinetic energy (KE), vorticity, and viscous energy loss (vEL). However, gradient-based metrics (vorticity, vEL) are highly sensitive to partial volume effects near the fluid-tissue boundary. This study investigated the impact of systematic contour boundary offsets on these parameters to standardise analysis. Five cases underwent 4D flow imaging. Deep learning-derived automated segmentations of the cardiac chambers were generated. Haemodynamics were analysed using three contouring methods: the baseline mask, a one-voxel inward offset, and a two-voxel inward offset. KE, vorticity, and vEL decreased progressively with larger offsets. KE declined modestly with erosion (by approximately 18% and 35% at one- and two-voxel offsets, respectively), a reduction commensurate with the loss of integration volume rather than the removal of boundary artefacts. By contrast, the gradient-based metrics were disproportionately sensitive to boundary proximity. In the left ventricle, mean full-cycle vorticity decreased from 249.6 ± 79.9 s<sup>-1</sup> (baseline) to 157.0 ± 60.4 s<sup>-1</sup> (two-voxel offset; Hedges' g 2.11), whilst vEL decreased from 549.4 ± 303.0 µW to 351.3 ± 230.0 µW (Hedges' g 2.00). A one-voxel inward offset optimally reduces boundary noise for sensitive gradient-based parameters. While KE analysis remains satisfactory using unmodified baseline contours, we recommend the uniform application of a one-voxel offset across all parameters to ensure methodological simplicity and pipeline standardisation.