Adaptive Morph-Patch Transformer for Aortic Vessel Segmentation

TL;DR

The paper tackles the challenge of accurately segmenting the aorta and its branches, where fixed-patch Transformers struggle to preserve complex vascular topology. It introduces the Adaptive Morph-Patch Transformer (MPT), which uses a velocity-field guided Morph-Patch strategy to produce morphology-aware patches and a Semantic Clustering Attention (SCA) to fuse features across semantically similar patches. The approach achieves state-of-the-art performance across three public datasets (AVT, TBAD, AortaSeg24) in 2D and 3D settings, with notable gains in Dice, mIoU, and clDice metrics, and demonstrates strong topology preservation, particularly for slender vessels. These results indicate robust, topology-aware segmentation with potential clinical impact for cardiovascular diagnostics and planning.

Abstract

Accurate segmentation of aortic vascular structures is critical for diagnosing and treating cardiovascular diseases.Traditional Transformer-based models have shown promise in this domain by capturing long-range dependencies between vascular features. However, their reliance on fixed-size rectangular patches often influences the integrity of complex vascular structures, leading to suboptimal segmentation accuracy. To address this challenge, we propose the adaptive Morph Patch Transformer (MPT), a novel architecture specifically designed for aortic vascular segmentation. Specifically, MPT introduces an adaptive patch partitioning strategy that dynamically generates morphology-aware patches aligned with complex vascular structures. This strategy can preserve semantic integrity of complex vascular structures within individual patches. Moreover, a Semantic Clustering Attention (SCA) method is proposed to dynamically aggregate features from various patches with similar semantic characteristics. This method enhances the model's capability to segment vessels of varying sizes, preserving the integrity of vascular structures. Extensive experiments on three open-source dataset(AVT, AortaSeg24 and TBAD) demonstrate that MPT achieves state-of-the-art performance, with improvements in segmenting intricate vascular structures.

Figures (6)

• Figure 1: Transformer improvements for tailoring vascular structure segmentation. (a) Traditional Transformers create fixed-size patches. (b) Our method creates morphed patches. (c) Morphed patches are grouped based on semantic similarities.
• Figure 2: Overall architecture of the proposed Morph-Patch Transformer for vascular structure segmentation. (a) Morph partition block to create morphology-aware patches. (b) The Transformer block with spatial and semantic attention. (c) Spatial and semantic attention is combined by window attention and semantic clustering attention.
• Figure 3: Accuracy–Efficiency Comparison on TBAD: (a) FLOPs, (b) Parameters; Colors Indicate Backbone Types.
• Figure 4: Visual results on the AVT and TBAD datasets. The first two rows illustrate the aortic segmentation performance of different models on the AVT dataset, while the bottom two rows present their ability to identify aortic dissection on the TBAD dataset. Red-boxed areas have been magnified to better visualize specific anatomical structures.
• Figure 5: Comparison of different models in processing complex vascular structures on the AortaSeg24 dataset. Red boxes denote incomplete anatomical structures, green boxes highlight significant noise, and yellow boxes indicate misclassifications.