<p>Timely treatment and improved patient outcomes are dependent on early recognition of breast cancer. Even though a variety of deep learning methods have been utilized in preceding studies, their ability to identify breast cancer has frequently been limited. This research presents Breast Cancer Detection from Mammograms, a unique method. To get around these obstacles, an Optimized Multi-Instance Gates-Controlled Deep Unfolding Network (MGCDUN–WOA) is used. The method is evaluated using the Inbreast and DDSM datasets. It utilizes Hybrid Fast Conventional Bilateral Filtering for preprocessing images, which improves quality by eliminating noise and artifacts. An efficient Cascaded Graph Convolutional Network is applied for accurate segmentation of breast cancer areas. Multi-Instance Gates-Controlled Deep Unfolding Network (MGCDUN) handles feature extraction and classification, by the Multi-Head Transformation (MHT) method extracting salient features. For the classification of features for breast cancer diagnosis, the GCDUNet is employed. Further, for enhancing the performance of the model, the Walruses Optimization Algorithm (WOA) is applied to optimize performance along with maximizing robustness and efficiency. The MGCDUN–WOA method provides excellent results, obtaining a recall rate of 99.8% and accuracy of 99.9% on the Inbreast and DDSM datasets, better than current methods and providing great potential to enhance early-stage breast cancer recognition and patient treatment.</p>