Deep learning-assisted 3D CT angiography for supply-vessel localization in inflammatory peripheral pulmonary artery pseudoaneurysms.

Inflammatory peripheral pulmonary artery pseudoaneurysms (PAP) are rare but potentially fatal causes of massive hemoptysis. Identifying the true parent/supply vessel before intervention is challenging, particularly when bronchial-pulmonary (B-P) shunts and dual systemic pulmonary supply is present. Digital subtraction angiography (DSA) is often invasive and exploratory. We developed a deep learning-assisted 3D-CT angiography (3D-CTA) workflow for localizing PAP supply vessels and compared its findings to DSA. The workflow includes V-Net-based segmentation of the PAP and pulmonary arterial tree from chest CTA, followed by 3D reconstruction, interactive vessel tracing, and geometry verification. Retrospective evaluation in six patients with inflammatory peripheral PAP (angiographic subtypes A-C) confirmed by CTA and DSA showed that all six datasets were successfully reconstructed into 3D models. Segmentation achieved a mean Dice similarity coefficient (DSC) of 0.884 for PAP and 0.932 for the PA tree. Supply-vessel localization was fully concordant with DSA in four cases and partially concordant in two cases. Given the substantial procedural burden of DSA (mean duration 212 min), the proposed 3D-CTA roadmap can streamline endovascular planning and potentially reduce the need for extensive DSA procedures. This pilot study provides preliminary evidence supporting the feasibility of a deep learning-assisted 3D-CTA workflow for PAP supply-vessel localization. Given the small sample size, these findings should be interpreted cautiously, and further validation in larger cohorts is required before clinical generalization.