AENN: An Integrated Attention-Enhanced Neural Network Architecture for Automated Histopathology Image Assessment

Histopathological images are a vital source of information in detecting cancer at an early stage, clinical diagnosis, and treatment planning. Nevertheless, the conventional convolutional neural networks (CNNs) are prone to fail to observe the fine spatial organizations and contextual relationships of the high-resolution whole slide images (WSIs). In order to overcome these difficulties, this paper suggests an Attention-Enhanced Neural Network (AENN) that combines a dual-attention system with a Transformer-based contextual embedding model to achieve better histopathology image classification. The model employs channel and spatial attention through the use of the Convolutional Block Attention Module (CBAM) on top of an EfficientNet-B3 backbone and then Transformer encoders to learn long-range inter-patch attention. The results of extensive experiments on the BreakHis dataset with different magnifications prove that AENN performs better, with the binary classification accuracy of 96.4, precision of 95.6, recall of 96.1, and F1-score of 95.8. In the multi-class cases, the model is able to maintain very high scores in benign and malignant subtypes. Multi-scale feature fusion and feature classification along with the hierarchical classification further improves performance and is still computationally efficient. Such findings suggest that AENN is a clinically sound, interpretable, and highly generalizable system of automated histopathology analysis.