DICNet: A Novel CNN Model based on DenseNet with Interleaved Convolutional Block Attention Module for the Classification of Breast Cancer Histopathology Images

Histopathology image analysis is crucial for the accurate diagnosis of breast cancer (BC), which is the most prevalent cancer among women. The prompt secondary opinions based on automated analysis can assist pathologists, reduce workload, and minimize interobserver variability. This paper introduces the DICNet, a novel convolutional neural network (CNN) architecture based on DenseNet with interleaved convolutional block attention module (CBAM) to classify the BC histopathology images. Specifically, the interleaved attention mechanism utilizes inter-channel and inter-spatial relationships among convolutional features to accentuate more informative and discriminative features. Furthermore, the pooled responses from multiple convolutional layers with varying spatial resolutions are concatenated to obtain a multi-scale representation. The classification performance of the proposed model is evaluated on the publicly available benchmark BreaKHis dataset with four magnification levels. The proposed DICNet model achieves an overall best classification accuracy of $94.02 \%$ at the $40 \times$ magnification level. Additionally, DICNet demonstrates enhanced classification performance at every magnification level, and significantly outperforms existing state-of-the-art approaches.