Enhancing Breast Cancer Detection: A CycleGAN-Enhanced Approach with Enhancing Sparrow Search Optimization

Breast cancer remains one of the leading causes of mortality among women worldwide, necessitating the development of highly accurate and computationally efficient diagnostic methods. Recent advancements in artificial intelligence and deep learning have provide the way for more reliable automated systems to assist in tumor classification. In this study, a novel hybrid classification framework called Cycle-consistent Adversarial Adaptation Network with Enhancing Sparrow Search Optimization (Cy2AN-ESSO) is proposed to improve the diagnostic performance for breast cancer detection using two benchmark datasets: BreakHis and TCGA-BRCA. The proposed approach begins with pre-processing using Robust Double-Weighted Guided Image Filtering (RD-WGIF) to enhance image quality and remove noise, ensuring that relevant features are preserved for subsequent stages. For segmentation, the Memory Efficient Vision Transformer (MEVis-Tr) is employed to accurately identify and isolate tumor regions, providing a reliable foundation for feature extraction. Feature extraction and classification uses the Multimodal Cycle-Triplet Adaptation Attention Network (MCT2AN), which integrates Triplet Attention Network (TAN) for feature extraction and the Cycle-consistent Adversarial Adaptation Network (C-2AN) for robust classification. The process is optimized using Enhancing Sparrow Search Optimization (ESSO) to refine decision boundaries and improve learning accuracy. The experimental results demonstrate that Cy2AN-ESSO demonstrated improved classification performance than current techniques for measuring accuracy and precision as well as recall and sensitivity and specificity and F1-score. Both accuracies and precisions and recalls reach 98.4%, 98.6% and 98.8% respectively for BreakHis data as well as TCGA-BRCA subtypes IDC, ILC, DCIS and LCIS. The proposed framework provided better classification accuracy and computational efficiency over the test datasets. More validation is needed on independent clinical cohorts before clinical use. Using Cy2AN-ESSO enables robust breast cancer classification that shows strong potential for accurate early disease diagnosis.

Hali tarjima qilinmagan