Binary Classification of Breast Tumours using CBAM-Enhanced Collaborative Network

For women breast cancer is a life-threatening problem, making the importance of early detection crucial. Expert pathologists assess cell shapes in breast histopathology images at various magnification levels for accurate diagnosis. Unfortunately, a shortage of pathologists in many countries heightens the risk of human error in disease identification and treatment. To address this challenge, this study proposed a CBAM-Enhanced Collaborative Network for classifying benign and malignant tumors in breast cancer histopathological images, offering a potential solution to enhance accuracy and efficiency in diagnosis. The model included a transfer learning branch using a pretrained DenseNet121 structure with integrated CBAM (Convolutional Block Attention Module). Simultaneously, a collaborative branch employed Inception-ResNet-v2 with CBAM for feature extraction from histopathological images. The feature fusion module then optimized information from both branches. The proposed method achieved impressive binary classification accuracies on the BreaKHis dataset: 98.31% overall, with 98.33%, 98.08%, 99.67%, and 97.80% for magnification factors of 40x, 100x, 200x, and 400x, respectively. CBAM-Enhanced Collaborative Network also exhibits strong generalization capabilities, as evidenced by its superior performance on the ICIAR2018 BACH Challenge dataset.