An Efficient CNN with Multiscale Feature Extraction for Breast Cancer Histopathology Analysis
Abstract
Timely and accurate diagnosis of breast cancer plays a vital role in ensuring effective treatment outcomes. Histopathology image analysis is reliable, while time-consuming and prone to human error due to the large number of slides that pathologists must review. To address these challenges, we propose a computationally efficient Convolutional Neural Network (CNN) architecture for classifying breast cancer histopathology images into benign and malignant categories. The model leverages repeated Depthwise-Multiscale-Selective-Kernel-Pointwise (DM-SKP) blocks to capture multiscale spatial features and employs a shallow-deep feature fusion strategy to retain both low-level and high-level information. Data augmentation balances the dataset and reduces overfitting. Extensive experiments on the BreakHis dataset demonstrate that the proposed approach achieves high performance, with accuracy reaching 91.16% and AUC of 0.98. Ablation studies confirm the effectiveness of the DMSKP blocks, and hyperparameter analysis highlights optimal learning settings. The results indicate that the proposed model provides a robust, accurate, and computationally efficient solution for automated breast cancer histopathology classification.