Panoramic Visual Summaries for Efficient Reading of Capsule Endoscopy Videos

Wireless capsule endoscopy (WCE) is a minimally invasive procedure enabling visual examination of the entire gastrointestinal tract. It is performed by a swallowable camera-equipped capsule wirelessly transmitting color video frames. Reading the resulting video is a challenging process for the endoscopists. It requires a considerable amount of time, usually hours of undivided attention and experience to detect and correctly interpret findings. In this paper we propose a novel method for automatic construction of panoramic visual summaries of WCE video frames enabling efficient reading of the video, while providing a broader field of view for the endoscopists. It involves extraction of visual features from each frame, and frame registration. Consecutive frames are geometrically transformed and stitched together to form new sequences of panoramic frames, each of which summarizes the information contained in multiple original ones. The proposed method has been integrated with Java Video Analysis (JVA), a novel software development framework for video applications. Its experimental evaluation on real, publicly available WCE video clips show that its application can result in a reduction of 84.5% of the video frames to be examined by the endoscopists.