Abnormalities Classification in WCE Images Using Pretrained Deep Learning Networks

Abstract

Wireless capsule endoscope (WCE) is the only device that provides endoscopic images of the entire small intestine. However, because of the large number of images it produces, the physicians need to spend a long time reviewing them, for the diagnosis. Therefore, Automatic computer-aided diagnosis tools are highly needed, to reduce the burden of physicians. In this paper, we present an automatic classification of different lesions in WCE images. We propose a deep learning method based on an ensemble of two pre-trained convolutional neural networks (CNN), namely MobileNet and DenseNet169. The extracted features from the entire selected architectures are concatenated and then fed into a multilayer perceptron, for the classification task. Experimental results proved that the proposed method improves the performance compared to the individual CNN classifiers.