Classification of Breast Cancer Histopathological Images with Deep Transfer Learning Methods

Abstract

It is very important to have a high accuracy rate in detecting cancerous cells in histopathological images. Thanks to high-accuracy images, cancerous cells will be detected more sensitively, and there will be a chance for more accurate and early diagnosis. Thus, a very important preliminary step will be taken in the treatment of cancerous cells. In this study, classification performances were comparatively analyzed by applying various methods to four different cancer cell types (benign, normal, carcinoma in situ and invasive carcinoma). By using BACH and Bioimaging as datasets, the desired parts are tried to be obtained primarily by several image processing methods (pyramid mean shifting, line detection, spreading). After obtaining images of different sizes, their performances are examined by using VGG16, DenseNet121, ResNet50, MobileNetV2, InceptionResNetV2, CNN deep transfer learning methods.