OCCNET: Improving Imbalanced Multi-Centred Ovarian Cancer Subtype Classification in Whole Slide Images

Ovarian carcinoma is known for its diverse subtypes with unique morphologies and clinical characteristics, causing considerable diagnostic complexities. While deep learning has emerged as a promising avenue for subtype identification, the intrinsic imbalance in different datasets from multiple medical institutions is still a challenging issue. In this paper, we introduce a novel Ensemble Attention Mechanism (EAM) specifically tailored to enhance subtype identification within imbalanced, multicentered ovarian cancer Whole Slide Images (WSIs), namely OCCNet. The OCCNet framework incorporate ensemble learning principles and attention mechanisms to adaptively balance the feature representation and effectively integrate every representation which mitigates inherent imbalances of class distribution. Then, one publicly available extensive Multi-centered Dataset from the ongoing Kaggle Competition is used to evaluate OCCNet's performance. Results show that the proposed method achieved a 93.67% F1-Score and a 93% Balanced Accuracy, revealing its feasibility and superiority.