EndoVLG: Region-to-Region Vision-Language Modeling for Endoscopy Report Generation.
[NEW PUBLICATION]
Congratulations to teacher Trang Mai Xuan and members Khanh Truong Long, Tung Nguyen Vu, Cong Nguyen Chi for having their research article officially accepted for publication at the SOICT 2026 International Conference!
Proceedings / Publisher:Springer CCIS
Conferences:SOICT 2026 (International Symposium on Information and Communication Technology)
Article title:
EndoVLG: Region-to-Region Vision-Language Modeling for Endoscopy Report Generation.
Main contributions:
Regional Visual-Linguistic Modeling: Proposing the EndoVLG architecture that combines multi-branch information, helping to accurately link the lesion area on endoscopic images and the corresponding medical text description.
Optimal multi-branch feature extraction for endoscopic images: Integrated region decoder/encoder, combining automatic segmentation and geometry extraction to fully capture surface and spatial details of the lesion.
Multimodal Alignment & Automatically generate diagnostic reports: Use the Multimodal Alignment mechanism combined with a large language model to automatically generate detailed, accurate medical diagnostic reports, supporting doctors during the endoscopic image diagnosis process.