Wolf Mayfly Optimization-driven Dense Xception Network for Enhanced Mammogram-based Breast Cancer Diagnosis

Breast cancer is a malignant tumor that most commonly originates in various regions of the breast, such as the lobules, connective tissue, and milk ducts. The screening and detection of breast cancer pose significant challenges in computer vision because of the existence of small yet clinically important irregularities in breast tissue growth. Existing models often exhibit poor performance in detecting small and low-contrast tumor regions in mammogram images. Many approaches rely on single-scale feature extraction, which restricts their ability to effectively capture both fine-grained and high-level semantic features. In this research work, a new technique, namely the Dense Xception Net Wolf Mayfly Optimization Algorithm (DXp-Net_WMyOA), is established for enhanced mammogram-based breast tumor diagnosis. Initially, the input image is preprocessed by the Medav filter for noise reduction. Afterwards, the Multi-Scale Feature Fusion Network (MSF-Net) is exploited for the segmentation of the cancer region. Thereafter, image augmentation is carried out based on random erasing, super pixel-mixing, rotation, and shifting, and then, feature extraction is accomplished. Further, the extracted features are forwarded to the Dense Xception Net (DXp-Net) for breast tumor detection. The DXp-Net is constructed by integrating the Xception network and DenseNet, and the Wolf Mayfly Optimization Algorithm (WMyOA) is utilized for training the DXp-Net. The WMyOA is introduced by integrating the Mayfly Algorithm (MA) and Wolf Bird Optimization (WBO). Moreover, the model achieved superior values of accuracy, sensitivity, specificity, MCC, Kappa, and F1-score of 94.879%, 95.467%, 94.989%, 93.566%, 91.657%, and 91.874%. These results indicate the effectiveness of the proposed approach for reliable and accurate detection of breast cancer.