Optimized Subtype Classification of Epithelial Ovarian Cancer using Augmented Vision Transformer Network

Epithelial Ovarian Carcinoma (EOC) is a critical global health issue, with accurate subtype classification being essential for effective patient treatment. Current methods often lack precision, leading to suboptimal outcomes. To address this, the study proposes the Augmented ViT_16 model, which integrates a complex fully connected network at the output stage of the standard ViT_16 encoder to enhance the classification of EOC subtypes, including high-grade serous, clear-cell, endometrioid, low-grade serous and mucinous carcinomas. Utilizing histopathological images, the proposed approach achieved 96% accuracy, with high precision (0.96), recall (0.95) and F1-score (0.96). By analyzing attention filters from the final encoder block, the study provides insights into the model’s decision-making process. The results demonstrate the potential of Vision Transformers in improving accuracy and robustness in EOC subtype classification, contributing to better targeted treatment strategies.