Return-Aligned Random Graph Diffusion with Dual-Channel Temporal Convolutional Network-Based Classification of Epithelial Ovarian Cancer on T2W-MRI

This study aims to develop a highly accurate and efficient deep-learning framework for the automated classification of epithelial ovarian cancer (EOC) subtypes using T2-weighted MRI (T2W-MRI) images. The objective is to overcome limitations such as poor contrast, high inter-class variation, dataset imbalance, and computational complexity that hinder current diagnostic methods. To address these, we propose the return-aligned random graph diffusion with dual-channel temporal convolutional network (RA-RGD-DCTCNet) model, evaluated on the Gene Expression Omnibus (GEO) and The Cancer Genome Atlas (TCGA) datasets. Image quality is first enhanced using Discrete Wavelet Transformation with Pre-Gaussian Filtering (DWT-PGF), followed by precise tumor segmentation via the return-aligned decision transformer (RADT). The random graph diffusion dual-channel temporal convolutional network (RGD-DCTCNet) performs feature extraction and classification, with accuracy further boosted by the Secretary Bird Optimization Algorithm (SBOA). Experimental results demonstrate that the RA-RGD-DCTCNet model achieves 99.9% accuracy and 99.8% sensitivity, significantly outperforming existing methods and showing promise for clinical application in reliable, automated cancer diagnosis.