Deep Learning Models for Aorta Segmentation in Computed Tomography Images: A Systematic Review And Meta-Analysis

Abstract

Purpose This systematic review and meta-analysis was conducted to evaluate the usefulness of deep learning (DL) models for aorta segmentation in computed tomography (CT) images. Methods Adhering to 2020 PRISMA guidelines, we systematically searched PubMed, Embase, and Web of Science for studies published up to March 13, 2024, that used DL models for aorta segmentation in adults’ chest CT images. We excluded studies that did not use DL models, involved nonhuman subjects or aortic diseases (aneurysms and dissections), or lacked essential data for meta-analysis. Segmentation performance was evaluated primarily in terms of Dice scores. Subgroup analyses were performed to identify variations related to geographical location and methodology. Results Our review of 16 studies indicated that DL models achieve high segmentation accuracy, with a pooled Dice score of 96%. We further noted geographical variations in model performance but no significant publication bias, according to the Egger test. Conclusion DL models facilitate aorta segmentation in CT images, and they can therefore guide accurate, efficient, and standardized diagnosis and treatment planning for cardiovascular diseases. Future studies should address the current challenges to enhance model generalizability and evaluate clinical benefits and thus expand the application of DL models in clinical practice.