Overview of ImageCLEFmedical 2026 – Medical Concept Detection and Caption Generation with Synthetic Data Extensions, CLEF working notes, Jena, Germany, 2026.
| Tipo de publicação: | Artigo |
| Citação: | |
| Journal: | Coeur |
| Volume: | 4283 |
| Ano: | 2026 |
| Mês: | October |
| Páginas: | 3122-3145 |
| Nota: | Actes de conférence: CLEF 2026 Working Notes |
| Resumo: | The 10th Edition of the ImageCLEFmedical Caption task extends the established concept detection and caption prediction benchmark with two synthetic-data subtasks and a manually evaluated explainability task. The 2026 release is built from finalized 2025 outputs and expanded with new PubMed Central data and updated annotations. The main dataset contains 131853 images (97364 train, 19240 validation, 15249 test) and 2646 unique UMLS Concept Unique Identifiers (CUIs). The synthetic-caption dataset contains 131723 images (97222 train, 19 239 validation, 15262 test) and 2174 unique CUIs. Both datasets were harmonized through missing-image pruning, zero-concept removal, and CUI-to-name mappings. The final test phase received 102 finished runs across four prediction leaderboards, and the explainability task received three manually evaluated submissions. Best hidden-test scores were 0.5790 F1-score for main concept detection, 0.3755 overall for main caption prediction, 0.5609 F1-score for synthetic concept detection, and 0.5840 overall for synthetic caption prediction. The best explainability submission reached an overall manual score of 3.69 on a 1–5 scale. This paper documents the 2026 task design, data preparation, synthetic-data extension, evaluation setup, organizer baselines, and final results |
| Palavras-chave: | computer vision, Image Captioning, ImageCLEF, Multi-label classification, Radiology, Synthetic Data, Vision-Language Models |
| Autores | |
| Adicionado por: | [] |
| Total mark: | 0 |
|
Anexos
|
|
|
Notas
|
|
|
|
|
|
Tópicos
|
|
|
|
|
