Answers are grounded in this paper and cite their sources.
Deep learning study demonstrates high-fidelity contrast-free virtual CT angiography synthesis using non-contrast scans, highlighting a safer alternative to iodinated contrast agents.
Computed tomography angiography (CTA) is a cornerstone of head and neck vascular assessment, but it requires the administration of iodinated contrast agents, which can cause adverse effects and are contraindicated in vulnerable patients. To enable contrast-free angiographic visualization, deep learning-based approaches have been proposed, which generate virtual CTA (vCTA) from non-contrast CT (NCCT). However, most existing methods adopt a single, global synthesis strategy, which often fails to delineate thin, spatially sparse vessels and degrades background anatomy. In this paper, we propose a novel localization-to-synthesis framework, called L2S-CTA. In the first stage, a localization network predicts a vascular map that localizes vasculature-related regions from NCCT. To promote vessel continuity, we introduce an inter-slice local attention (ILA) that aggregates contextual features from neighboring axial slices. In the second stage, we synthesize vCTA conditioned on the predicted localization prior, allowing the synthesizer to focus on vasculature-related regions. Experiments on both internal and external cohorts support the feasibility of the proposed framework for retrospective vCTA synthesis. The source code will be available at https://github.com/keunsoo-ko/L2S-CTA .
No takes yet. Share an insight, caveat, or question.
Ko et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: