Breast Cancer Classification Based on CNN and ESIM Model
Yang Lu, Zequn Zhao, Ziyu Zhao · 2021 IEEE International Conference on Computer Science, Electronic Information Engineering and Intelligent Control Technology (CEI) · 2021
Breast cancer is a major cause of death among women worldwide, and invasive ductal carcinoma (IDC) is the most common form of breast cancer. However, the recognition of images of hematoxylin and eosin (H&E) stained breast histopathology samples has the limitations of low efficiency and low accuracy in previous studies. Also, there is a lack of network platforms that can automatically interact with patients. To tackle these issues, this paper proposed a deep learning-based method for the classification of H&Estained breast tissue images released from Andrew Janowczyk's website by fine-tuning the VGG convolutional neural network (CNN). Moreover, the enhanced LSTM for natural language inference (ESIM) model is also proposed to process text matching in website intelligent question answering. As a result, we can achieve 80% accuracy on the validation set regarding image recognition, which significantly improves upon a previous benchmark. Furthermore, in terms of text matching, the ESIM model achieves 74% accuracy on the verification set, which meets the website's requirements for intelligent questions and answers. Therefore, patients can interact with the website that combines CNN and ESIM models to quickly obtain accurate preliminary breast cancer diagnosis results.