Classifying digestive organs in wireless capsule endoscopy images based on deep convolutional neural network

This paper studies the classification problem of the digestive organs in wireless capsule endoscopy (WCE) images based on deep convolutional neural network (DCNN) framework. Essentially, DCNN proves having powerful ability to learn layer-wise hierarchy models with huge training data, which works similar to human biological visual systems. Classifying digestive organs in WCE images intuitively means to recognize higher semantic image features. To achieve this, an effective deep CNN-based WCE classification system has been constructed (DCNN-WCE-CS). With about 1 million real WCE images, intensive experiments are conducted to evaluate its performance by setting different network parameters. Results illustrate its superior performance compared to traditional classification methods, where about 95% classification accuracy can be achieved in average. Moreover, it is observed that the DCNN-WCE-CS is robust to the large variations of the WCE images due to the individuals and complex digestive tract circumstance, including the rotation, the luminance change of the WCE images.