Design and Development of Hybrid Learning Model for the detection of Ovarian Cancers

Abstract

Ovarian cancer remains as a significant contributor to female mortality across the globe, mainly because it has a difficulty in obtaining an early and precise diagnosis through complex histopathology images. Automated image-based diagnosis has therefore become essential to assist pathologists in making effective diagnoses of malignant tissues. Traditional diagnostic procedures and Deep Learning (DL) architectures often fail to capture fine-grained cellular textures as well as global contextual relationships needed for accurate classifications. These models are based on narrow receptive fields, have no focus on critical areas, and thus, the diagnostic accuracy and generalization across subtypes are limited. In order to overcome these shortcomings, the present research proposes a Hybrid Deep Learning Framework (ViT-SE + BiGRU-Luong FusionNet) to detect ovarian cancer accurately and effectively. The Vision Transformer (ViT) with a Squeeze-and-Excitation (SE) module is used to learn rich global contextual dependencies and focus important feature channels adaptively. Simultaneously, the Bidirectional Gated Recurrent Unit (BiGRU) with the Luong Attention mechanism accurately captures the sequential spatial dependencies and identifies critical pathological areas in tissue sequences. Finally, a Cross-Attention Fusion (CAF) module is utilised to combine complementary global and sequential representations to achieve robust classification. The proposed framework is trained and tested on the Kaggle Ovarian Cancer and Subtypes Histopathology Dataset and achieves an accuracy of 99.53%, precision of 99.41%, recall of 99.47%, specificity of 99.45% and F1-score of 99.44%. The experimental findings prove that the proposed hybrid framework substantially improves diagnostic accuracy and generalization without compromising high computational efficiency, which makes it very applicable in real-time clinical practice in medical image diagnosis.