Three-Dimensional Reconstruction and Geometric Characterization of Diseased Coronary Arteries: Comparing Visual Estimation and 3D-QCA Against FFR

Coronary artery disease (CAD) is a major global health problem, and clinical decision making is often based on anatomical severity of the stenosis and its functional relevance. In practice, however, anatomy and physiology do not always align, complicating the decision-making. In this thesis a centerline-based geometric characterization was developed, computing curvature, tortuosity, and local radius, from 3D reconstructions of diseased coronary segments to look beyond simple lumen narrowing. Reconstructions were performed for 28 patients and integrated with 25 additional cases from a companion thesis, for a total of 53 coronary reconstructions on which standardized geometric descriptors were computed. The aim of this work was to examine how anatomical and functional severity are related. Two anatomical assessment approaches were evaluated: (a) physician visual estimation (VE), reflecting experience-driven practice and a qualitative reading of the angiogram, and (b) software-based Three-Dimensional Quantitative Coronary Angiography (3D-QCA), which uses calibrated edge detection and reference-diameter modeling to compute percent diameter/area stenosis and lesion length. Functional relevance of stenoses was assessed exclusively with fractional flow reserve (FFR), the wire-based ratio of distal coronary to aortic pressure under maximal hyperemia. In this study, the analysis quantified the association between FFR and two anatomical properties of the coronary lesion, percentage stenosis and lesion length, each measured by physician visual estimation (VE) and by automated 3D-QCA. VE-based percent stenosis showed a weak but significant inverse correlation with FFR (r = &#8722;0.3440, p-value = 0.0117). Also, VE-derived lesion length demonstrated a weak, significant, inverse correlation with FFR (r = &#8722;0.3969, p = 0.0033), in line with fluid-dynamic expectations that longer constrictions increase hydraulic losses, generating greater pressure drops and thus FFR decreases. As for automated quantification, 3D-QCA percent stenosis yielded a moderate inverse correlation with FFR (r = &#8722;0.4911, p = 0.0002), indicating that calibrated edge detection and reference-diameter modeling improve the coupling between anatomy and physiology relative to VE. In contrast, the 3D-QCA lesion-length metric did not correlate significantly with FFR (r = &#8722;0.1665, p = 0.2333). In general, VE overestimates percentage stenosis, compared with 3D-QCA (median VE 60% vs 3D-QCA 46%, p-value < 0.0001), whereas 3D-QCA overestimates lesion length compared with VE (median VE 10.0 mm vs 3D-QCA 22.2 mm, p < 0.0001). These biases suggest that (a) algorithmic edge-based diameter estimates temper physicians’ tendency to upstage severity, and (b) automated centerline/path-based length definitions may include shoulder or taper regions that inflate lesion extent beyond visually adjudicated segments. Taken together, these findings clarify where anatomy does and does not track physiology. The results motivate explicit calibration between VE and 3D-QCA in routine workflows, emphasize quality control for automated length estimation, and support the integration of centerline-based geometric features to account for residual variance in FFR unexplained by diameter loss alone. This work provides a reproducible template for interpreting discordant anatomy–physiology cases, guiding invasive decision-making and informing non-invasive planning and lays the groundwork for future models that combine standardized geometry with validated physiological endpoints.