DS4DH group at ImageCLEFmedical caption 2025 : Notebook for the DS4DH Lab at CLEF 2025

Article Article scientifique Article DS4DH group at ImageCLEFmedical caption 2025 : Notebook for the DS4DH Lab at CLEF 2025 eng 2025-09-22 author He Jiawei 444500 Département de radiologie et informatique médicale (RADIO) author Ferdowsi Sohrab 0000-0003-3768-6408 787220 Département de radiologie et informatique médicale (RADIO) author Feng Weibo 431332 Département de radiologie et informatique médicale (RADIO) author Alves Fernando author Platon Alexandra 554893 Département de radiologie et informatique médicale (RADIO) author Teodoro Douglas 0000-0001-6238-4503 682380 Département de radiologie et informatique médicale (RADIO) Douglas Henrique Teodoro 2448-2460 1613-0073 https://ceur-ws.org/Vol-4038/ This paper presents the DS4DH team’s approaches for the ImageCLEFmedical Caption 2025 challenge, where we participated in two tasksConcept Detection and Caption Prediction. For Concept Detection, unlike the typical approach of multi-label classification, we posed the problem as image to sequence of tokens mapping, where the tokens consist of the CUI’s, and a standard transformer maps the image embeddings into the sequence. Our approach achieved an F1-score of 52.3%, ranking top-6 among the best submission systems. For Caption Prediction, we developed multiple approaches, including fine-tuned InstructBLIP, traditional retrieval-augmented generation (RAG), and cluster-based RAG methods. Our best strategy, based on the InstructBLIP model, achieved the highest recall (BERTScore (Recall) of 60.7%) among all participants, being ranked top-2 according to the Overall challenge metric (33.6%). Our experiments reveal that RAG approaches did not outperform the baseline, exposing critical challenges in medical image captioning where noisy retrievals significantly weaken generation performance. Through validation experiments and case studies, we demonstrate that only highly accurate reference images prove helpful, as poor retrieval quality introduces noise that degrades caption generation. Presented at CLEF 2025, held in Madrid, Spain, 9-12 September 2025. ImageCLEF RAG Image Embedding Radiology Dewey 616.0757 62 Département de radiologie et informatique médicale 227 false 1379 DS4DH - Data Science for Digital Health 1035 469 Radiologie des urgences 800 CEUR workshop proceedings 4038 d574c73c-21a2-4f2d-8c20-838c2c30cf01 Article (Published version) Public paper_191.pdf 1886731 application/pdf CC BY-4.0 ec73cb6c932282f30bde9976847c8835fbd7d108ecce7992782669a71ca780f5 openAccess Faculté de médecine / Section de médecine clinique / Département de radiologie et informatique médicale

Title	Published in	Access level	OA Policy	Year	Views	Downloads

DS4DH group at ImageCLEFmedical caption 2025 : Notebook for the DS4DH Lab at CLEF 2025	CEUR workshop proceedings			2025	18	205

Archive ouverte UNIGE

He, Jiawei