CV-Attention UNet: Attention-based UNet for 3D Cerebrovascular Segmentation of Enhanced TOF-MRA Images

TL;DR

This work targets automated cerebrovascular segmentation in TOF-MRA, addressing challenges of small vessel visibility and limited data by introducing CV-AttentionUNet, a 3D-UNet backbone augmented with spatial attention and deep supervision. The pipeline includes Hessian-based Frangi vessel enhancement and a data-labeled/unlabeled dual-dataset setup, enabling robust learning from limited data. Key contributions are the attention gate with group normalization to handle small-batch training, and the use of deep supervision to fuse multi-scale features, resulting in state-of-the-art performance on TubeTK data (DSC around 70.9% on labeled data and up to 91.7% on extended labels) with high specificity and competitive Hausdorff distances. The approach demonstrates promise for end-to-end 3D cerebrovascular segmentation on TOF-MRA, potentially aiding stroke diagnosis and pre-surgical planning, though future work should address disease-specific data, channel attention, and augmentation strategies.

Abstract

Due to the lack of automated methods, to diagnose cerebrovascular disease, time-of-flight magnetic resonance angiography (TOF-MRA) is assessed visually, making it time-consuming. The commonly used encoder-decoder architectures for cerebrovascular segmentation utilize redundant features, eventually leading to the extraction of low-level features multiple times. Additionally, convolutional neural networks (CNNs) suffer from performance degradation when the batch size is small, and deeper networks experience the vanishing gradient problem. Methods: In this paper, we attempt to solve these limitations and propose the 3D cerebrovascular attention UNet method, named CV-AttentionUNet, for precise extraction of brain vessel images. We proposed a sequence of preprocessing techniques followed by deeply supervised UNet to improve the accuracy of segmentation of the brain vessels leading to a stroke. To combine the low and high semantics, we applied the attention mechanism. This mechanism focuses on relevant associations and neglects irrelevant anatomical information. Furthermore, the inclusion of deep supervision incorporates different levels of features that prove to be beneficial for network convergence. Results: We demonstrate the efficiency of the proposed method by cross-validating with an unlabeled dataset, which was further labeled by us. We believe that the novelty of this algorithm lies in its ability to perform well on both labeled and unlabeled data with image processing-based enhancement. The results indicate that our method performed better than the existing state-of-the-art methods on the TubeTK dataset. Conclusion: The proposed method will help in accurate segmentation of cerebrovascular structure leading to stroke

Figures (9)

• Figure 1: The overall diagram for 3D cerebrovascular segmentation using the deep learning model.
• Figure 2: Preprocessing steps: From left to right, the steps are bias correction, skull stripping, Hessian based vessel enhancement and 3D patch generation.
• Figure 3: Proposed cerebrovascular attention UNet architecture (CV-AttentionUNet) and deep supervision.
• Figure 4: Diagram for attention module.
• Figure 5: Detailed training, validation and testing steps.