Implementasi Metode CNN Multi-Scale Input dan Multi-Feature Network untuk Dugaan Kanker Payudara

Abstract

According to WHO, cancer is one type of disease with a high increase in terms of cases around the world. Breast cancer is the highest contributor to morbidity rates in 2020, which is 2.26 million cases. In determining the patient's prognosis, several examinations are needed, one of them is histopathological analysis. However, histopathological analysis is a relatively tedious and time-consuming process. With the development of deep learning, computer vision can be applied for detection in medical images, which is expected to help improve the accuracy of the prognosis and the speed of identification carried out by experts. Based on this knowledge, this study aims to implement multi-class classification (normal, benign, in situ, invasive) and prediction of normal digital tissue images or has suspected cancer cells using the Convolutional Neural Network with multi-scale and multi-feature network (CNN-G). The dataset used is 400 breast tissue image data which are classified into four classes and labeled by a pathologist. The accuracy result obtained from the training is 0.5375~0.54 and has made an increase when the result was compared to single models (CNN14, CNN42, CNN84). Other model evaluation methods conducted are confusion matrix, precision, recall, and f-1 score.