Selective sampling bias in coronary CT quantification: mechanism and evaluation with TRACE
Under Consideration at Biomedical Signal Processing and Control (BSPC)
Abstract
Rejecting unreliable vessel sections changes the anatomical sample used to quantify coronary stenosis. Accuracy measured only on retained sections can therefore miss sampling bias. We developed an evaluation framework using the Traceable Representation of Anatomy and Coronary Evidence (TRACE) to link measurements, sampling positions and assessment states. Three measurement configurations were compared in 256 computational volumes. Exact geometric references decomposed profile error into retained-section measurement error and spatial-selection bias. Equal-count random and circular-shift controls tested sampling location. A separate 192-volume experiment crossed calcium location with geometric twist. Clinical reprocessing used 46 retrospective development-cohort patients, with nine candidate-support settings evaluated on fixed measurements. Image-unclipped measurement reduced diameter mean absolute error from 1.051 to 0.079 mm relative to mask guidance. In the coupled calcium/twist context, selection contributed –25.00 percentage points to a total profile bias of –28.59. In 24 positive-severity blocks, moving calcium from outside to over the stenosis shifted the exact retained-profile descriptor by –16.78 points at zero twist (95% confidence interval –20.74 to –12.79). The effect persisted at 40°. Program-defined clinical whole-path support fell from 142/184 to 6/184 branches. Image-unclipped capture at 50% was 0/24 angiography-reference-positive branches in the 14-patient report-reconciled subset. Selective loss of narrow sections can thus distort stenosis estimates despite low local measurement error. Evaluating measurement error, sampling location and endpoint availability together identifies this failure mechanism and separates local-positive evidence from whole-path assessment.
Ethics Statement
The retrospective analysis of institutional clinical data was approved by the Ethics Committee of the First Affiliated Hospital with Nanjing Medical University (Jiangsu Province Hospital), Nanjing, China (approval no. 2025-SR-1242; 7 January 2026). The study was conducted in accordance with the Declaration of Helsinki. The requirement for informed consent was waived by the approving ethics committee. Results are reported without directly identifying patient information.
