Two new research papers introduce advanced deep learning models for segmenting coronary arteries in Digital Subtraction Angiography (DSA) images. The first, HTC-SGA Former, utilizes a hybrid Transformer-CNN architecture with a novel Boundary-Weighted Adaptive Compound Loss (BWACL) to improve the segmentation of thin, low-contrast vessels and their boundaries. The second, MSA-UNet3+, employs a Multi-Scale Attention UNet3+ framework combined with a Supervised Prototypical Contrastive Loss (SPCL) to address class imbalance and enhance feature differentiation for more precise vessel delineation. Both methods demonstrate superior performance compared to existing state-of-the-art techniques on private datasets, offering more reliable analysis for cardiovascular interventions. AI
IMPACT These models offer improved accuracy in segmenting coronary arteries, potentially leading to more precise diagnosis and treatment planning for cardiovascular diseases.
RANK_REASON Two academic papers published on arXiv detailing new deep learning models for medical image segmentation.
- Boundary-Weighted Adaptive Compound Loss
- CNN
- coronary artery disease
- Coronary DSA
- HTC-SGA Former
- MSA-UNet3+
- Rayan Mahjoub Merghani Ahmed
- Rayan Merghani Ahmed Mahjoub
- Supervised Prototypical Contrastive Loss
- transformer
- UNet3+
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
