Reproduce “CIS-UNet: Multi-Class Segmentation of the Aorta in Computed Tomography Angiography via Context-Aware Shifted Window Self-Attention” on OpenResearch

About this paper

Muhammad Imran, Jonathan R Krebs, Veera Rajasekhar Reddy Gopu, Brian Fazzone, Vishal Balaji Sivaraman, Amarjeet Kumar, Chelsea Viscardi, Robert Evans Heithaus and 4 more

Researchers from the University of Florida and UC Santa Cruz developed CIS-UNet, a deep learning model for multi-class 3D segmentation of the aorta and 13 major branches from CTA images. The model achieved a Dice Similarity Coefficient of 0.713 and Mean Surface Distance of 2.767 mm, outperforming state-of-the-art methods while offering faster inference for complex surgical planning.

Key insights

• The strategic placement of the Context-aware Shifted Window Self-Attention (CSW-SA) module solely at the bottleneck layer allows for efficient capture of global contextual information, overcoming limitations of local window-based attention without significantly increasing computational burden.
• A hybrid architecture that effectively combines CNNs for local feature extraction with tailored Transformer blocks for global dependency modeling can achieve superior accuracy and continuity in complex 3D medical image segmentation tasks.
• Detailed multi-class segmentation of the aorta and numerous individual branches is achievable with deep learning, providing a fully automated and precise tool for advanced endovascular surgical planning that outperforms existing methods.

Summary by alphaXiv.