2026· Conference and Labs of the Evaluation Forum· pp. 3463-3477· 1 citation· 30 references
Computer Science
TL;DR
Large vision-language models can produce useful medical image captions when combined with soft concept-level guidance and careful output validation, however, the gap between high semantic similarity and lower MedCAT/alignment scores indicates that exact biomedical grounding remains challenging.
Results indicate that a carefully optimized ImageNet-initialized DenseNet-121 with imbalance-aware learning provides an effective and reliable solution for extreme long-tail medical concept detection.
T. Q. Tô, D. C. Pham, N. H. Bui et al.· Conference and Labs of the E...· 1 citation
A replicable two-stage pipeline for clinical GI VQA combining parameter-efficient fine-tuning with deterministic output normalisation is proposed and an empirical characterisation of branch-ordering sensitivity and context-gate fragility in rule-based post-processors under distribution shift is characterised.
Tan N. Nguyen, Bao N. Huynh-Dang, Thien B. Nguyen-Tat· Conference and Labs of the E...· 1 citation
This paper presents the AUEB/DMST submission to the 2026 ImageCLEFmedical Caption Tasks, namely the Concept Detection and Caption Prediction Tasks, and implemented a captioning architecture with dual visual encoders and a single decoder, connected by a Q-Former-based cross-attention bridge.
Giorgos Karandreas, P. Louridas· Conference and Labs of the E...· 1 citation
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.