Fig. 3: Comparative strategies for tumor–vessel encasement angle assessment. | npj Digital Medicine

Fig. 3: Comparative strategies for tumor–vessel encasement angle assessment.

From: A clinically validated 3D deep learning approach for quantifying vascular invasion in pancreatic cancer

Fig. 3

A Manual 2D estimation on a single axial CT slice (α = 121°), simulating conventional radiologic evaluation. Clinicians typically select the slice with maximal apparent tumor–vessel contact and visually estimate the angle of encasement. Tumor and vessel masks are shown in red and cyan, respectively. This method is constrained to a single imaging plane. B Semi-automated 3D measurement using conventional software (θ = 154.4°). Radiologists place three reference points (P0–P2) along the tumor–vessel interface to compute the encasement angle in three-dimensional space. Although this approach allows multi-planar assessment, the measurement relies on manually selected reference points, which may be influenced by user input variability. C Fully automated angle quantification using PAN-VIQ (β = 172.4°). The model automatically identifies the region of maximal tumorvessel interaction and calculates the encasement angle on a reconstructed 3D plane. Two perspectives are shown to illustrate geometric consistency. Tumor is rendered in red, vessels in green.

Back to article page