Exploring Deep Architectures for Accurate Breast Cancer Prediction using Histopathological Images

Breast cancer continues to be a leading cause of morbidity and mortality among women globally. While histopathological analysis of biopsy specimens remains the clinical standard for diagnosis, it is often labor-intensive and prone to subjective interpretation. This work presents a comparative evaluation of two advanced convolutional neural network (CNN) models—EfficientNet-B0 and DenseNet121— applied to the BreakHis dataset across four magnification levels: 40x, 100x, 200x, and 400x. Following an extensive review of existing methodologies and a detailed explanation of the training strategy and interpretability methods with a focus on GradCAM, we report that EfficientNet-B0 and DenseNet121 achieve mean classification accuracies of 99.72% and 98.71%, respectively, across magnifications. These results establish new performance baselines for automated breast cancer detection from histopathological images. This also assists oncologists to initiate appropriate treatment for the betterment of patients.