![[2602.15138] MB-DSMIL-CL-PL: Scalable Weakly Supervised Ovarian Cancer Subtype Classification and Localisation Using Contrastive and Prototype Learning with Frozen Patch Features](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.15138] MB-DSMIL-CL-PL: Scalable Weakly Supervised Ovarian Cancer Subtype Classification and Localisation Using Contrastive and Prototype Learning with Frozen Patch Features
Summary
This paper presents a novel approach for classifying and localizing ovarian cancer subtypes using weakly supervised learning techniques, achieving significant improvements in accuracy while maintaining scalability.
Why It Matters
Ovarian cancer diagnosis and treatment personalization are critical, yet current methods face scalability challenges. This research introduces an innovative AI approach that enhances classification accuracy without compromising efficiency, potentially transforming pathology practices.
Key Takeaways
- The proposed method improves F1 scores by 70.4% and 15.3% for instance- and slide-level classification, respectively.
- Utilizes contrastive and prototype learning with frozen patch features, enhancing scalability in training.
- Achieves significant AUC gains for instance localization (16.9%) and slide classification (2.3%).
- Addresses the increasing diagnostic workloads in pathology departments with AI solutions.
- Highlights the shift from traditional methods to more efficient end-to-end feature extraction techniques.
Computer Science > Computer Vision and Pattern Recognition arXiv:2602.15138 (cs) [Submitted on 16 Feb 2026] Title:MB-DSMIL-CL-PL: Scalable Weakly Supervised Ovarian Cancer Subtype Classification and Localisation Using Contrastive and Prototype Learning with Frozen Patch Features Authors:Marcus Jenkins, Jasenka Mazibrada, Bogdan Leahu, Michal Mackiewicz View a PDF of the paper titled MB-DSMIL-CL-PL: Scalable Weakly Supervised Ovarian Cancer Subtype Classification and Localisation Using Contrastive and Prototype Learning with Frozen Patch Features, by Marcus Jenkins and Jasenka Mazibrada and Bogdan Leahu and Michal Mackiewicz View PDF HTML (experimental) Abstract:The study of histopathological subtypes is valuable for the personalisation of effective treatment strategies for ovarian cancer. However, increasing diagnostic workloads present a challenge for UK pathology departments, leading to the rise in AI approaches. While traditional approaches in this field have relied on pre-computed, frozen image features, recent advances have shifted towards end-to-end feature extraction, providing an improvement in accuracy but at the expense of significantly reduced scalability during training and time-consuming experimentation. In this paper, we propose a new approach for subtype classification and localisation in ovarian cancer histopathology images using contrastive and prototype learning with pre-computed, frozen features via feature-space augmentations. Compared to DSMIL, our met...
