A magnification-independent method for breast cancer classification using transfer learning

Abstract

Breast cancer is the most common and diagnosed cancer among women worldwide. Doctors use breast imaging reporting and data systems to classify breast density. This study proposes a transfer learning-based artificially intelligent (AI) system to classify breast cancer from the histopathological images of the breast. Three different Deep Convolutional Neural Networks (DCNN) architectures, namely, the Visual Geometry Group-16 (VGG-16), Depthwise Separable Convolutions (Xception), and Dense Convolutional Network-201 (Densenet-201), have been used as a base model in the transfer learning method used in this work. Each test image has been classified into benign or malignant class after extracting the features from each test image using the three pre-trained base models. The proposed classification system is independent of the magnification factors of the images. The performance of the proposed breast cancer classification system has been evaluated on two widely used public datasets known as the Invasive Ductal Carcinoma (IDC) dataset and BreaKHis dataset. The classification accuracies of 99.42% (IDC dataset) and 99.12% (BreaKHis dataset) have been obtained from the proposed system. The results demonstrate that the proposed breast cancer classification system outperforms the state-of-the-art methods in this research domain.

Keywords

• Transfer learning
• Deep convolutional neural networks
• Artificial intelligence
• Breast cancer classification
• Histopathologic images