An interactive prediction system of breast cancer based on ResNet50, chatbot and PyQt

Abstract

Breast cancer has gradually become an important killer that endangers people’s health. How to diagnose breast cancer quickly and accurately has become a popular research direction. However, traditional testing by the doctor is time-consuming and laborious, and there is still the problem of accuracy. Deep learning becomes a tool of evidence-based medicine, which can effectively solve the above problems and realize the function of detecting breast cancer automatically and with high accuracy. In our study, we selected and applied the optimal CNN model named ResNet50 for breast cancer diagnosis. Due to the small size of images in our dataset, the 3*3 convolutional layer performed better than the 7*7 convolutional layer in our breast cancer classification task. Besides, our pre-trained ResNet50 achieved 94.698% accuracy on the WSI dataset, while un-pretrained ResNet50 only achieved 93.777%. The result presented pretraining in the ImageNet dataset can more effectively reduce the loss and improve accuracy. We also developed an application that integrated the CNN model with a chatbot implemented by NLTK and an interface constructed through PyQt.