Diameter measurement of tubular structures on CT angiography, 3D rotational angiography, and 2D digital subtraction angiography against caliper ground truth: A phantom study of surrogates for intracranial vessels.

BACKGROUND AND PURPOSE: Measured intracranial vessel dimensions guide device selection in cerebrovascular intervention, yet accuracy depends on imaging modality, luminal contrast concentration, and the lumen-edge detection criteria. Prior accuracy studies benchmark one angiographic modality against another rather than physical ground truth. This study quantifies accuracy of CT angiography (CTA), 3D rotational angiography (3DRA), and 2D digital subtraction angiography (DSA) against a phantom of tubular contrast-filled structures serving as surrogates for intracranial vessels. MATERIALS AND METHODS: Phantoms with cylindrical lumen diameters 1.2-8.5 mm at graded iodinated contrast concentrations (6.25-100%) were imaged with clinical devices in air and a tissue-equivalent gel surround. Tubular cross-sections perpendicular to centerlines were measured by maximum-gradient, full-width-half-maximum (FWHM), and fixed-Hounsfield criteria and compared with physical phantom measurements using mixed-effects models. RESULTS: Across 1682 cross-sections (73 lumens), all modalities recovered true diameter within ~0.5 mm but consistently measured larger than the true dimensions. Over-measurement occurred minimally in the 2-5 mm range, maximally at the extremes of caliber and contrast concentration. 3DRA was modestly closer to physical phantom measurements than CTA (gradient bias +0.40 vs +0.60 mm). Maximum-gradient edge over-measured diameters versus the conventionally preferred FWHM, with the latter carrying larger, more concentration-sensitive bias. Fixed-HU thresholds were concentration-biased on CTA and unusable on 3DRA. 2D DSA, currently the clinical reference standard, was itself biased in analysis of its auto-calibrated measurements, over-measuring large lumens, under-measuring small lumens 1.5 mm, and reading larger frontally than laterally through magnification. CONCLUSIONS: Conventional CTA, 3DRA, and 2D DSA each measure intracranial-scale caliber lumens within ~0.5 mm but with consistent bias. Maximum-gradient lumen edge detection minimizes bias and outperforms FWHM, while fixed-Hounsfield contour tracing is unreliable, particularly on 3DRA. Characterizing these error modes provides a calibration basis for automated vessel-measurement platforms.