Deep Learning Models for Classification of Cancer Data Using Pre-Trained CNN Based Architectures

Abstract

The visual examination of histopathological images is the gold standard in the diagnosis of breast cancer, but an intricate and strenuous process that takes years of pathologist training. Hence, automating this task using computer assisted modelling is highly anticipated. This article suggests an approach of transfer learning from the histopathological images for automated classifications of breast cancer disease. Five different convolutional neural networks (CovNets) are used for the prediction of breast cancer disease. VGG16 net, Mobile net, Resent50, Resent101, and Resent152 are used for the prediction of the disease. The loss and accuracy of the trained model and the validation data set is performed. Using confusion matrix visualization technique, the precision, F2 score, recall, and support are thoroughly analyzed for each model separately. The analysis revealed that VGG16 net and Resent152 are capable of predicting the breast cancer disease conveniently compared to other models. Resent50 failed to predict the disease appropriately.