Ovarian cancer remains a leading cause of cancer-related mortality among women worldwide. Traditional diagnostic methods often lack the precision required for early detection and accurate subtype classification. In this study, we address the challenge of automating ovarian cancer diagnosis by introducing Attention-Based Models (ABMs) in combination with 3D Convolutional Neural Networks (CNNs). Our research seeks to enhance the accuracy and efficiency of ovarian cancer diagnosis, particularly in distinguishing between serous, mucinous, and endometrioid subtypes. Conventional diagnostic approaches are limited by their reliance on manual interpretation of medical images and fail to fully exploit the rich information present in MRI scans. The proposed work leverages ABMs to dynamically focus on critical regions in MRI scans, enabling enhanced feature extraction and improved classification accuracy. We demonstrate our approach on a well-curated dataset, OvaCancerMRI-2023, showcasing the potential for precise and automated diagnosis. Experimental results indicate superior performance in cancer subtype classification compared to traditional methods, with an accuracy of 94% and F1 score of 0.92. Our findings underscore the potential of ABMs and 3D CNNs in revolutionizing ovarian cancer diagnosis, paving the way for early intervention and more effective treatment strategies. In conclusion, this research marks a significant advancement in the realm of ovarian cancer diagnosis, offering a promising avenue for improving patient outcomes and reducing the burden of this devastating disease. The integration of ABMs and 3D CNNs holds substantial potential for enhancing the accuracy and efficiency of ovarian cancer diagnosis, particularly in subtyping, and may contribute to early intervention and improved patient care.