Classification of Breast Cancer Based on Histopathological Image Using EfficientNet-B0 on Convolutional Neural Network

Abstract

Breast cancer has been chosen as the leading cause of cancer-related death in women. Biopsy is still the most accurate way to detect cancer cells. However, this is time-consuming and requires a relatively expensive cost and requires a pathologist. Advances in machine learning make it possible to detect and diagnose breast cancer using histopathological images that are the result of a biopsy. BreakHis dataset is a dataset that provides histopathological images. This study proposes the use of this dataset for cancer classification based on histopathological images using EfficientNet-B0 on the Convolutional Neural Network (CNN). The purpose of this study is to improve the performance of previous studies and to determine the effect of augmentation and dropout layers on the proposed model. This study also applies cross-validation to get more accurate results. The results showed that the EfficientNet-B0 model combined with augmentation and dropout layers with the application of k-fold cross validation k=7 was able to improve performance in previous studies with an accuracy of 98.90%. The application of data augmentation and dropout layer techniques has also been shown to increase accuracy and reduce overfitting of the proposed model.