Convolutional neural network with parallel convolution scale attention module and ResCBAM for breast histology image classification

• DOI: 10.1016/j.heliyon.2024.e30889
• Publication Date: 2024-05-08T07:55:53Z
• Authors: Ting Yan; Guohui Chen; Huimin Zhang; Guolan Wang; Zhenpeng Yan; Ying Li; Songrui Xu; Qichao Zhou; Ruyi Shi; Zhi Tian; Bin Wang
• Keywords: Convolution (computer science); Histology; Contextual image classification

Breast cancer is the most common cause of female morbidity and death worldwide. Compared with other cancers, early detection breast more helpful to improve prognosis patients. In order achieve diagnosis treatment, clinical treatment requires rapid accurate diagnosis. Therefore, development an automatic system for suitable patient imaging great significance assisting treatment. Accurate classification pathological images plays a key role in computer-aided medical prognosis. However, recognition methods images, scale information, loss image information caused by insufficient feature fusion, enormous structure model may lead inaccurate or inefficient classification. To minimize impact, we proposed lightweight PCSAM-ResCBAM based on two-stage convolutional neural network. The included Parallel Convolution Scale Attention Module network (PCSAM-Net) Residual Convolutional Block (ResCBAM-Net). first-level was built through 4-layer PCSAM module prediction patches extracted from images. optimize network's ability represent global features tiled fusion method fuse patch same image, residual attention module. Based above, second-level constructed predictive We evaluated performance our ICIAR2018 dataset BreakHis dataset, respectively. Furthermore, ablation studies, found that dilated convolution play important improving performance. Our outperforms existing state-of-the-art models 200 × 400 magnification datasets maximum accuracy 98.74 %.