An Intra- and Cross-frame Topological Consistency Scheme for Semi-supervised Atherosclerotic Coronary Plaque Segmentation

TL;DR

This work addresses coronary plaque segmentation from CTA CPR frames, where boundary ambiguity and limited annotations hinder performance. It introduces ICTC, a dual-consistency semi-supervised framework combining Intra-frame Topological Consistency (ITC) and Cross-frame Topological Consistency (CTC). ITC uses a dual-task network to predict segmentation $\hat{S}$ and Skeleton-aware Distance Transform $\hat{R}$, enforcing topology via $L_{ITC}$ without extra annotations; CTC employs an unsupervised flow estimator $\hat{O}_{f}$ to enforce cross-frame continuity through $L_{CTC}$. Across CAA-Seg, AD-CTA, and ACDC datasets, ICTC outperforms other semi-supervised methods and approaches supervised performance, with ablations confirming the contributions of the SDT branch and both topological constraints. The approach demonstrates strong generalization and suggests a practical path for leveraging unlabeled data in precise medical image segmentation tasks, particularly where topology and frame-wise continuity are informative priors.

Abstract

Enhancing the precision of segmenting coronary atherosclerotic plaques from CT Angiography (CTA) images is pivotal for advanced Coronary Atherosclerosis Analysis (CAA), which distinctively relies on the analysis of vessel cross-section images reconstructed via Curved Planar Reformation. This task presents significant challenges due to the indistinct boundaries and structures of plaques and blood vessels, leading to the inadequate performance of current deep learning models, compounded by the inherent difficulty in annotating such complex data. To address these issues, we propose a novel dual-consistency semi-supervised framework that integrates Intra-frame Topological Consistency (ITC) and Cross-frame Topological Consistency (CTC) to leverage labeled and unlabeled data. ITC employs a dual-task network for simultaneous segmentation mask and Skeleton-aware Distance Transform (SDT) prediction, achieving similar prediction of topology structure through consistency constraint without additional annotations. Meanwhile, CTC utilizes an unsupervised estimator for analyzing pixel flow between skeletons and boundaries of adjacent frames, ensuring spatial continuity. Experiments on two CTA datasets show that our method surpasses existing semi-supervised methods and approaches the performance of supervised methods on CAA. In addition, our method also performs better than other methods on the ACDC dataset, demonstrating its generalization.

Figures (4)

• Figure 1: CPR Schematic. There is spatial continuity in the frame images, and their topological structure can be used as prior information.
• Figure 2: The overview of our method: (a) ITC Pipeline. Two adjacent frame images share a dual-task network, involving regression and segmentation tasks. We simultaneously construct ITC between the regression results and SDT labels, between the regression results and segmentation results. (b) CTC Pipeline. Input and split the regression results into a single one. Via an unsupervised flow estimator. CTC is constructed on the adjacent regression results with the learned pixel flow.
• Figure 3: Visualization of the results on the CAA-Seg. In the figure, the red is the lumen and the white is the calcified plaque.
• Figure 4: Visualization of the results on the ACDC. In the figure, the red is the right ventricle (RV), the blue is the left ventricle (LV), and the green is the myocardium.