Breast Cancer Classification from Histopathological Images with Inception Recurrent Residual Convolutional Neural Network

Abstract

The Deep Convolutional Neural Network (DCNN) is one of the most powerful and successful deep learning approaches. DCNNs have already provided superior performance in different modalities of medical imaging including breast cancer classification, segmentation, and detection. Breast cancer is one of the most common and dangerous cancers impacting women worldwide. In this paper, we have proposed a method for breast cancer classification with the Inception Recurrent Residual Convolutional Neural Network (IRRCNN) model. The IRRCNN is a powerful DCNN model that combines the strength of the Inception Network (Inception-v4), the Residual Network (ResNet), and the Recurrent Convolutional Neural Network (RCNN). The IRRCNN shows superior performance against equivalent Inception Networks, Residual Networks, and RCNNs for object recognition tasks. In this paper, the IRRCNN approach is applied for breast cancer classification on two publicly available datasets including BreakHis and Breast Cancer (BC) classification challenge 2015. The experimental results are compared against the existing machine learning and deep learning-based approaches with respect to image-based, patch-based, image-level, and patient-level classification. The IRRCNN model provides superior classification performance in terms of sensitivity, area under the curve (AUC), the ROC curve, and global accuracy compared to existing approaches for both datasets.

Keywords: Breast cancer recognition; Computational pathology; DCNN; Deep learning; IRRCNN; Medical imaging.

Fig. 1

Implementation diagram for breast cancer recognition using the IRRCNN model. The upper part of this figure shows the steps that are used for training the system, and the lower part of this figure displays the testing phase where the trained model is used. These results are evaluated with a number of different performance metrics

Fig. 2

Diagrams displaying the inception recurrent residual unit (IRRU) consisting of the inception unit and recurrent convolutional layers that are merged by concatenation, and the residual units (summation of the input features with the outputs of the inception unit can be seen just before the output block)

Fig. 3

The first row shows the four types of benign tumors, and the second row shows the malignant tumors. The magnification factor of these images is ×400

Fig. 4

Sample images of four types of breast cancer (normal, benign, in situ carcinoma, and invasive carcinoma) from the 2015 BC Classification Challenge dataset

Fig. 5

Four example images with corresponding augmented images. The actual images are shown on the left, and four augmented samples (of the 20 created for each image) are shown on the right

Fig. 6

Center patch and resized images from an original sample (left) and from an augmented sample (right)

Fig. 7

Training and validation accuracy for BC classification with 8 classes for the IRRCNN model at different magnification factors

Fig. 8

ROC curve with AUC for different magnification factors for eight class BC classification

Fig. 9

Training and validation accuracy for the multi-class case using the 2015 BC Classification Challenge dataset. Sample sets are resized and augmented (RZ + AUG), center patch cropped and augmented (CRP + AUG), random patches (RP), sample resized (RZ), or center patch cropped (CRP)