Retrieval-Augmented, Vision-Language Framework for Ovarian Carcinoma Subtype Classification
Cancer Informatics · 2026-11 · 25
DOI: 10.1177/11769351261452669 ↗Kurz gefasst
Fachartikel in der Zeitschrift Cancer Informatics, online erschienen am 18.11.2026. Verfasst von Vidip Gaur und 2 weiteren Autor:innen. Die Arbeit nennt 29 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Purpose
Ovarian carcinoma subtype classification is critical for prognosis and treatment planning but remains challenging due to subtle morphological differences and limited expert availability. This study aims to develop an interpretable and computationally efficient framework for accurate subtype identification from whole-slide images.
Methods
We propose a lightweight, retrieval-based approach built on the CONCH vision-language model. High-resolution image patches are converted into descriptive captions, from which structured, class-specific phrases rep- resenting recurring histopathological patterns are extracted and stored in a searchable index. During inference, captions from unseen samples are embedded and matched to this index using cosine similarity. A rank-based scoring strategy quantifies feature overlap across subtypes, followed by semantic retrieval and frequency-based voting for final classification. A large language model is employed only for post hoc explanation generation and does not influence predictions.
Results
Experiments conducted on a publicly available ovarian carcinoma dataset demonstrate that the proposed method achieves performance comparable to existing deep learning-based approaches. The framework provides transparent decision-making while requiring substantially fewer computational resources and no end-to-end model training.
Conclusion
The proposed interpretable and lightweight retrieval-based frame- work enables accurate ovarian carcinoma subtype classification with enhanced transparency and efficiency. By decoupling prediction from explanation and avoiding heavy model training, the method offers a practical alternative for clinical deployment and supports trustworthy decision-making in pathology.
Zitieren
APA
Gaur, V., Gaur, V., & Mishra, T. K. (2026). Retrieval-Augmented, Vision-Language Framework for Ovarian Carcinoma Subtype Classification. Cancer Informatics, 25. https://doi.org/10.1177/11769351261452669
Vancouver
Gaur V, Gaur V, Mishra TK. Retrieval-Augmented, Vision-Language Framework for Ovarian Carcinoma Subtype Classification. Cancer Informatics. 2026;25. doi:10.1177/11769351261452669
BibTeX
@article{Gaur2026,
author = {Gaur, Vidip and Gaur, Vidu and Mishra, Tusar Kanti},
title = {Retrieval-Augmented, Vision-Language Framework for Ovarian Carcinoma Subtype Classification},
journal = {Cancer Informatics},
year = {2026},
volume = {25},
doi = {10.1177/11769351261452669}
}
EUVIMED