DATUM quantifies spatial support and reference effects in coronary diameter assessment
Under Consideration at Biomedical Signal Processing and Control (BSPC)
Abstract
Coronary diameter comparisons change when methods measure different parts of the arterial tree. Diameters of Arterial Trees through Unified Measurement (DATUM) demonstrates how anatomical reporting support, branch qualification and aggregation shape patient-level error contrasts. Four measurement configurations were evaluated on fixed predictions from two segmentation models for the same 160 ImageCAS patients in the ImageCAS-X test subset, using geometric references derived from public human-corrected segmentations. Under equal-branch weighting, switching from method-specific to shared reference-arc locations shifted error contrasts by more than 0.02 mm in 21.9–34.2% of patients across four branch-heuristic and VMTK-based comparisons. Branch composition and reweighting produced the larger mean shift under both tested weighting rules; within-branch position restriction acted in the opposite direction. Branch qualification amplified local interpolation changes: removing 0.046% of available grid values changed six branch-reportability decisions and shifted one patient's shared-location contrast by 0.129 mm. The separately tested interpolation and weighting alternatives retained all six negative shared-location mean contrasts, with changes in magnitude and individual signs. For FFR-U-Net, reference replacement reduced mean DATUM error by 0.01252 mm, compared with 0.00047 mm for readout replacement; all twelve crossed-construction mean contrasts favoured DATUM. These experiments identify the mechanisms linking local measurement availability to patient-level comparisons and quantify the influence of reference construction on reported error.
Keywords: Coronary arteries; Computed tomography; Diameter measurement; Spatial correspondence; Measurement support; Reproducibility;
