Bridged Semantic Alignment for Zero-shot 3D Medical Image Diagnosis

We propose a bridged semantic alignment framework combining explicit and implicit alignment strategies for zero-shot 3D medical image diagnosis. Addressing the challenge of limited 3D CT-report paired data, our implicit alignment mechanism optimizes cross-modal feature distributions to significantly improve semantic consistency and diagnostic generalization. Achieves SOTA performance on three open-source 3D CT benchmarks for zero-shot diagnosis and image-text retrieval.

Venue: IEEE Journal of Biomedical and Health Informatics — CAS Q1, JCR Q1 Top, IF: 6.8
Role: First Author

Recommended citation: Haoran Lai, et al. "Bridged Semantic Alignment for Zero-shot 3D Medical Image Diagnosis." IEEE Journal of Biomedical and Health Informatics, 2025.
Download Paper