Abstract
The human digestive system can be affected by many types of diseases. For example, three of the six most common cancer types (esophagus, stomach and colorectal) are located in the gastrointestinal tract. Colorectal cancer (CRC) is the third most common cancer in men and the second most common cancer in women worldwide, and Norway has one of the highest incidences of this cancer. Early detection is vital for the prognosis, level of treatment and survival. EIR is a multimedia system with the main objective of supporting doctors in gastrointestinal tract disease detection, both as a live examination system and an offline system for VCE. However, the detection and automatic analysis subsystem within EIR today consists of two parts; the detection subsystem and the localisation subsystem. Recent advances in machine learning, particularly deep learning, have provided excellent object detection models. This thesis explores the possibility of using a deep neural network at the base of the detection and automatic analysis subsystem in EIR, specifically by using You only look once (YOLO). YOLO is a state-of-the-art, real-time object detection system that was used together with the ASU Mayo Clinic polyp database to detect CRC precursors called polyps. The YOLO system reaches a satisfactory detection accuracy, while still being able to process videos in real-time. The proposed system and EIR is compared using the standard metrics of recall, precision and F1-score. When compared, it is clear that the system still has room for improvement in regard to its precision.