Utilising self-similarity to model the morphometry of the pulmonary arteries

This study aims to model the human pulmonary artery tree, addressing missing information in image segmentation through self-similarity, as demonstrated in pig lung studies. Current CT imaging models define artery tree geometry but often miss small lateral arteries, leading to errors in morphometric descriptions. Data from 10 never-smokers (5F/5M) with normal lung function were used. Intra-pulmonary vessel segmentation and artery/vein separation were performed using Chest Imaging Platform. Data were converted to 3D ‘graphs’ using Python KDTree functions. Sub-graphs were connected by recalibrating branches and connecting closest trees. Length to diameter ratio (L/D), child to parent diameter ratio (D/D_p), and branching angle (Θ) were calculated. Lateral branches were added to arteries with L/D > 5.0, with lateral dimensions set using values calculated for minor branches in the arterial tree (L/D = 2.46, D/D_p = 0.4) based on the assumption of self-similarity. Raw uncorrected data (N = 316 ± 65 branches) gave 3.57 ± 3.32, 0.86 ± 0.43, and 46.2 ± 32.7 for L/D, D/D_p, and Θ, respectively. Addressing connectivity gaps gave N = 330 ± 159 branches. Addition of lateral branches gave N = 452 ± 123 branches with metrics 2.89 ± 1.78, 0.80 ± 0.40, and 36.6 ± 31.97 for L/D, D/D_p, and Θ, respectively. This compares well with literature values of L/D = 2.8-3.25, D/D_p = 0.78-0.83, and Θ Utilising self-similarity to model the morphometry of the pulmonary arteries Θ = 15.3-57.0. Connecting disconnected sub-trees and supplementing with self-similar lateral branches yielded a plausible fully-connected tree with morphometry consistent with existing literature.