INTERACT-CMIL: Multi-Task Shared Learning and Inter-Task Consistency for Conjunctival Melanocytic Intraepithelial Lesion Grading

Ikinci M, Toma L, Loeffler KU, Ussem L, Susskind D, Weller J, Yeganeh Y, Herwig-Carl MC, Albarqouni S (2026)

Publication Type: Conference contribution

Publication year: 2026

Journal

Proceedings - International Symposium on Biomedical Imaging

Publisher: IEEE Computer Society

Book Volume: 2026-April

Conference Proceedings Title: Proceedings - International Symposium on Biomedical Imaging

Event location: London, GBR

ISBN: 9798331577636

DOI: 10.1109/ISBI61048.2026.11515389

Abstract

Accurate grading of Conjunctival Melanocytic Intraepithelial Lesions (CMIL) is essential for treatment and melanoma prediction but remains difficult due to subtle morphological cues and interrelated diagnostic criteria. We introduce INTERACT-CMIL, a multi-head deep learning framework that jointly predicts five histopathological axes; WHO4, WHO5, horizontal spread, vertical spread, and cytologic atypia, through Shared Feature Learning with Combinatorial Partial Supervision and an Inter-Dependence Loss enforcing crosstask consistency. Trained and evaluated on a newly curated, multicenter dataset of 486 expert-annotated conjunctival biopsy patches from three university hospitals, INTERACT-CMIL achieves consistent improvements over CNN and foundation-model (FM) baselines, with relative macro F1 gains up to 55.1%(WHO4) and 25.0% (vertical spread). The framework provides coherent, interpretable multicriteria predictions aligned with expert grading, offering a reproducible computational benchmark for CMIL diagnosis and a step toward standardized digital ocular pathology.

Authors with CRIS profile

Julia Weller Department of Ophthalmology

Involved external institutions

Universitätsklinikum Bonn DE Germany (DE) Technische Universität München (TUM) DE Germany (DE) Universitätsklinikum Tübingen DE Germany (DE)

How to cite

APA:

Ikinci, M., Toma, L., Loeffler, K.U., Ussem, L., Susskind, D., Weller, J.,... Albarqouni, S. (2026). INTERACT-CMIL: Multi-Task Shared Learning and Inter-Task Consistency for Conjunctival Melanocytic Intraepithelial Lesion Grading. In Proceedings - International Symposium on Biomedical Imaging. London, GBR: IEEE Computer Society.

MLA:

Ikinci, Mert, et al. "INTERACT-CMIL: Multi-Task Shared Learning and Inter-Task Consistency for Conjunctival Melanocytic Intraepithelial Lesion Grading." Proceedings of the 23rd IEEE International Symposium on Biomedical Imaging, ISBI 2026, London, GBR IEEE Computer Society, 2026.

BibTeX: Download