Ovarian cancer remains a major global health issue due to its late-stage diagnosis and the morphological similarity of its histological subtypes, which include Mucinous Carcinoma (MC), Endometrioid Carcinoma (EC), Low-Grade Serous Carcinoma (LGSC), Clear Cell Carcinoma (CCC), and High-Grade Serous Carcinoma (HGSC). Traditional histological assessment is constrained by subjectivity and inter-observer variability, prompting the development of automated, dependable diagnostic technologies. In this work, we present a hybrid deep learning architecture that combines a Convolutional Neural Network (CNN) for localized morphological feature extraction with a Vision Transformer (ViT) decoder to capture global tissue context. Using the UBC-OCEAN dataset, the proposed architecture achieved highly competitive results, with an AUC of 87.03 % and an accuracy of 86.45% across the main histological subtypes of ovarian cancer. These findings demonstrate the potential for hybrid CNN-Transformer techniques to support and improve diagnostic precision in clinical histopathology workflows.