Abstract

Automatic detection of diseases and anatomical landmarks in medical images by the use of computers is important and considered a challenging process that could help medical diagnosis and reduce the cost and time of investigational procedures and refine health care systems all over the world. Recently, gastrointestinal (GI) tract disease diagnosis through endoscopic image classification is an active research area in the biomedical field. Several GI tract disease classification methods based on image processing and machine learning techniques have been proposed by diverse research groups in the recent past. However, yet effective and comprehensive deep ensemble neural network-based classification model with high accuracy classification results is not available in the literature. In this thesis, we review ways and mechanisms to use deep learning techniques to research on multi-disease computer-aided detection about gastrointestinal and identify these images. We re-trained five state-of-the-art neural network architectures, VGG16, ResNet, MobileNet, Inception-v3, and Xception on the Kvasir dataset to classify eight categories that include an anatomical landmark (pylorus, z-line, cecum), a diseased state (esophagitis, ulcerative colitis, polyps), or a medical procedure (dyed lifted polyps, dyed resection margins) in the Gastrointestinal Tract. Our models have showed results with a promising accuracy which is a remarkable performance with respect to the state-of-the-art approaches. The resulting accuracies achieved using VGG, ResNet, MobileNet, Inception-v3, and Xception were 98.3%, 92.3%, 97.6%, 90% and 98.2%, respectively. As it appears, the most accurate result has been achieved when retraining VGG16 and Xception neural networks with accuracy reache to 98% due to its high performance on training on ImageNet dataset and internal structure that support classification problems.