A Comparative Study of Different Types of Convolutional Neural Networks for Breast Cancer Histopathological Image Classification
Farjana Parvin, Farjana Parvin, Md. Al Mehedi Hasan, Md. Al Mehedi Hasan
Abstract
Recently, for image classification and detection convolutional neural networks have shown breakthrough performance. The main reason behind their impressive performance is that they are inspired by the mammal's visual cortex. In this paper, we have investigated the performance of five convolutional neural network architectures that are LeNet-5, AlexNet, VGG-16, ResNet-50, and Inception-vl on the basis of test accuracy, AUC, precision, recall, f1-score. All of these models are tested on the BreaKHis dataset. In the BreaKHis dataset the Inception-vl network has achieved the highest test accuracy of 89%, 92%, 94% and 90% respectively at 40X, 100X, 200X and 400X magnification factor binary classification. The inception-v l network also has achieved the highest AUC of 85%, 91%, 94% and 88%, highest precision of 86%, 91%, 92% and 89%, highest recall of 85%, 91%, 94% and 88% and highest f1-score of 85%, 91%, 93% and 89% respectively at 40X, 100X, 200X and 400X magniflcation factor.