AI-Enhanced Mammographic Screening: Optimizing Breast Lesion Detection and Classification with the Robust ECA-Net152 Model

Breast cancer poses a significant threat to women’s health, and early screening and diagnosis are crucial for improving treatment outcomes and reducing mortality. Computer-aided diagnosis (CAD) systems have been developed to help physicians classify benign and malignant lesions in mammograms. The paper introduces a new pre-processing technique for mammographic datasets, addressing issues like tissue overlap and image noise. The technique includes denoising, segmentation, contrast adjustment, and data augmentation algorithms, enhancing CNN input and generalization. The ECA-Net152 model, which combines channel attention and focal loss function, is introduced to classify breast lesions. The method achieves an AUC of 0.960 and performs better through experimental comparisons and GradCAM visualization. This study contributes to AI-assisted diagnosis and early screening of breast cancer.