Automatic Classification of Breast Cancer Histopathological Images Based on a Discriminatively Fine-Tuned Deep Learning Model

Automatic classification of breast cancer histopathology images is of great significance in breast cancer diagnosis. Convolutional Neural Networks (CNN) requires the right hyper-parameters to efficiently and accurately do these classification, usually based on expertise and extensive trial and error, and also factored by the dataset. In this paper, the veracity of discriminative fine-tuned algorithm on ResNet and DenseNet Models, which optimally sets a range of hyper-parameters, using cyclical learning rate policy per iteration in training. The proposed method was tested on the Public BreakHis Dataset of Magnification 100X and 400X. Experimental results based on Accuracy metric (Densenet (100X): 95.33%, Densenet (400X): 94.34%, Resnet (100X): 96.56%, Resnet (400X): 96.3%) proves the method of optimization is efficient for breast cancer histopathology image classification in clinical settings.