Breast Cancer Image Recognition Based on a Lightweight DenseNet Model

Breast cancer remains a significant global threat to women's health, making early diagnosis and treatment vital for reducing mortality rates. Traditional methods for classifying breast cancer pathology images are time-consuming, labor-intensive, and often lack accuracy. While deep learning models have made remarkable strides in breast cancer image classification, they typically require extensive computational resources, including high-performance GPUs and large memory capacities, which can be prohibitive for many healthcare institutions. To address this challenge, this paper presents an enhanced DenseNet model that integrates Depthwise Separable Convolutions and Squeeze-and-Excitation blocks, aiming to improve classification performance while reducing computational costs. The proposed model is validated on the publicly available BreakHis dataset, which automatically classifies breast cancer pathology images into benign and malignant categories. Comparative experiments were conducted with other deep learning models, including VGGNet, DenseNet, ResNet, and MobileNetv2. The results demonstrate that the proposed model exhibits superior learning capability and achieves the best recognition performance, with an accuracy exceeding 99%, a recall of 99%, and a precision of 99% on this dataset.