A computer-aided method for the automatic localization of the left ventricular long axis on coronary computed...

The accurate identification of the left ventricular long axis is important for standardized cardiac orientation and quantitative analysis in coronary computed tomography angiography (CCTA). However, current workflows often require manual adjustment, leading to variability and increased workload. This study aimed to develop a computer-aided method for the automatic localization of the left ventricular long axis.

Eighty-eight patients undergoing CCTA were retrospectively analyzed. We employed a specific range of tangential slopes within the outer contour in the mid-to-lower region and applied further correction to build spatial localization of the long axis, with physician annotations on 3D Slicer serving as the reference. The consistency was assessed via the Spearman correlation coefficient and Bland-Altman analysis.

In the coordinate system, the physician-labeled long axis (Axis _a) and computer-measured long axis (Axis _d) formed the following median angles with the coronal and transverse planes, respectively: α _a, 104.10° [interquartile range (IQR), 98.95-111.92°]; α _d, 105.85° (IQR, 98.44-111.84°); and β _a, 126.03° (IQR, 120.89-133.78°); and β _d, 125.57° (IQR, 121.38-133.66°). Spearman analysis showed highly significant correlations at α and β (r=0.88 and r=0.87, respectively; P<0.001). Bland-Altman analysis demonstrated good agreement, with SDs of bias of 8.8 for α [95% limits of agreement (LoA): -16.98 to 17.65] and 4.62 for β (95% LoA: -9.49 to 8.60). For the comparison between the Axis _d and Axis _a, the median distance error was 19.42 (IQR, 10.25-36.70), and the median angular error was 4.40° (IQR, 2.36-8.66). These errors were comparable to the inter-reader differences between Axis _a and another physician-labeled long axis (Axis _b), with no significant difference (P>0.05). The inter- and intra-correlation coefficients were good [all intraclass correlation coefficients (ICCs) >0.9; P<0.001].

Determining and validating the left ventricular long axis from distinct ventricular features can markedly enhance the efficiency of image processing and analysis.