B60-30 AI-driven Assessment of the Distal Pulmonary Arterial Tree in Children Undergoing Single Ventricle Palliation Surgeries

Background

In infants with single-ventricle heart defects accurate pulmonary artery assessment is critical for guiding staged palliative procedures, yet current CT methods remain largely qualitative and limited for distal vasculature evaluation. AI-based segmentation may enable 3D visualization and quantitative analysis of distal pulmonary arterial architecture.

Objectives

To develop an automated AI-based method for pulmonary artery segmentation, generate 3D reconstructions, and evaluate whether quantitative CT-derived vascular characteristics can help to predict adverse surgical outcomes.

Methods

This single-center retrospective study included patients aged 0-2 years with single-ventricle heart defects who underwent chest CT angiography before and after stage 1 palliation (S1) between 2017 and 2025. Median time between CTs were 93 days. Pulmonary arteries were segmented using 3D Slicer using an artery-vein classifier independent of Hounsfield units via point-cloud-based representation of the vessel mask. Quantitative vessel tortuosity (QVT) measurements of vessel geometry and topology were extracted. Vessels were categorized based on diameter (<0.5 mm, 0.5-1 mm, 1-1.5 mm, >1.5 mm) and normalized to total vessel volume. Changes in QVT features obtained before and after S1 were compared. Mann-Whitney U and Spearman correlation were utilized.

Results

33 patients underwent staged palliation with pre- and post-operative CT scans available. Adverse outcomes included 4 deaths (12.1%), 1 transplant referral (3.0%), 4 unable to progress to next stage (12.1%), and 8 had shunt complications (24.2%). 18 patients (54.5%) had immediate Stage 1 complications, and 8 (26.7%) had Stage 2 complications; none occurred after Stage 3. Patients who required ECMO or developed shunt problems had higher ratios and volume of pulmonary arteries >1.5 mm; those requiring reintervention after Stage 1 had fewer <0.5 mm vessels and more 0.5-1.0 mm vessels. Patients associated with death, reintervention, and inability to progress to the next stage of palliation exhibited significantly decreased QVT-based distal pulmonary artery torsion, curvature, and geometric complexity on CT scans obtained before and after S1. Volumetric pulmonary artery changes before and after S1 were not associated with any adverse outcomes. Traditional markers (hospitalization, ICU stay, ventilation duration) were not significantly associated with outcomes.

Conclusions

Automated AI-based pulmonary artery segmentation enables quantitative assessment of distal vasculature and identifies vessel size patterns linked to adverse perioperative outcomes in single-ventricle patients. Quantitative CT metrics may provide unique prognostic information beyond conventional clinical variables and have potential to improve risk stratification and surgical planning. Multicenter validation of these measurements is warranted.

This abstract is funded by: None