Detection of Breast Cancer Using a Dual-Stream Network of DenseNet121 and U-Net Guided ViT Fusion Transformer

The development of effective breast cancer detection methods becomes essential because breast cancer continues to be the main reason for illness and mortality among female patients worldwide.This research adopts a dual-stream detection system that merges DenseNet121 features with ViT-Lite CNN segments that undergo late fusion integration.A pre-trained DenseNet121 extracts complex image features from breast ultrasound data as the first stream.In contrast, the second stream combines U-Net segmentation with ViT-Lite CNN to analyze tumor regions exclusively.The combined output from both streams enables the method to accurately classify benign, malignant, and normal cases.The proposed approach delivered 99,99% classification accuracy during evaluations using the Breast Ultrasound Images Dataset (BUSI), which reflects its strong effectiveness as a detection solution.The system design exhibits lightweight features with ease of scalability that match the requirements of resource-limited healthcare environments and telemedicine infrastructure.Data detection assessments indicate that combining convolutional neural networks and transformer modules offers improved potential for breast cancer diagnosis at an early stage, which benefits patient health outcomes.