Colitis detection in the gastrointestinal tract based on deep learning and machine learning

Approximately two million people worldwide die as a result of gastrointestinal disorders every year. Video endoscopy is one of the most recent medical imaging tools for diagnosing gastrointestinal illnesses such as polyps, colitis, and stomach ulcers. Because medical video endoscopy creates a large number of images, clinicians must devote significant time to reviewing all of them. This makes manual diagnosis difficult, which has led to research into computer-aided methods that can quickly and accurately diagnose all images that are generated. In this thesis, a system has been proposed to locate and type colon disease. The proposed system consists of two main phases which are (Detection, and Classification). The detection phase consists of preparing a dataset with two classes (normal, and abnormal), then extracting features using histogram orientation gradient (HOG) for each class then used these features by support vector machine (SVM) classifier for training detection phase. In the classification phase used a convolution neural network (CNN LeNet model) for training with four classes (Dyed-lifted-polyps, Esophagitis, Normal-cecum, and Ulcerative-colitis) depending on the results of SVM detection. The result archived from the detection phase for (Dyed lifted-polyps class, and Ulcerative-colitis class) was 98.33 %, and the accuracy of (Esophagitis class, and Normal-cecum class) was 96.67%. The average accuracy of detection was 97.5%. The prediction results of the Convolution neural network (CNN) was 100% for training and 95.38 for testing. Through proposed system it has been concluded that the use of (HOG) with SVM classifier it was very accurate and quick in locating the colitis. Also, using (CNN LeNet model) is better than other models for the dataset used.