Deep learning models for anomaly detection in wireless capsule endoscopy video frames: The transfer learning approach

Wireless Capsule Endoscopy (WCE) is a popular and widely accepted technique for the examination of gastro-intestinal (GI) tract and small bowel. A small capsule is swallowed by the patient equipped with a camera that records its journey in the form of a video. This video helps the doctor in visual examination of a patient’s GI tract and intestine which further helps in the diagnosis of diseases. Several image processing and machine learning techniques have been proposed by the researchers for abnormal frame detection from WCE videos. Recent approaches have used deep learning frameworks for abnormality detection. In this study, transfer learning is employed using various available deep learning models for classification of normal images and images with abnormality. Three popular models—InceptionV3, Resnet50, and InceptionResnetV2—were trained and compared. It is found that the performance deep learning models is quiet better in comparison to traditional machine learning methods and InceptionV3 outperformed with 93% accuracy.