Automatic Detection of Gastrointestinal Diseases Using Wireless Capsule Endoscopy Images

Gastrointestinal (GI) diseases are various disorders related to the digestive system. This system includes the esophagus, stomach, small and large intestines, liver, gallbladder and pancreas, starting from the mouth. Early diagnosis is very important in the treatment of the disease. The earlier the disease is diagnosed, the higher the chance of the patient being treated. In recent years, it is known that artificial intelligence techniques have been widely used in disease diagnosis and classification. Among the artificial intelligence techniques, deep learning methods that produce very successful results in image classification are frequently used. This success of deep learning methods has been tried to be used in the classification of GI diseases. Within the scope of this study, it was tried to detect bleeding GI or lesions from publicly available wireless capsule endoscopy (WCE) images. As a result of the experiments, 5 different deep learning architectures were used. Features were extracted from the two architectures that showed the highest accuracy and were combined. Neighborhood Component Analysis (NCA) dimension reduction method was applied to the obtained feature map and a hybrid model was obtained. It was seen that the proposed hybrid model achieved an accuracy value of 86.3%.