Capsule endoscopy is a medical diagnostic tool used to examine the digestive tract. It is actually a miniature camera that has its own energy source, LED light source and transmitter. After swallowing the capsule is conducted through the patient's gastrointestinal tract (esophagus, stomach, duodenum, small intestine, colon, rectum) in a span of about 8 hours, continuously taking pictures that are transmitted to be stored in a video container on a hard disk, fixed to the patient's waist. This examination method is quite effective, for example with this a significant part of intestinal tumors in more than 90% efficiency can be detected, whereas with other methods, this value is only about 10-15%. It is however not without its drawbacks: During an examination more than 50,000 images are generated, and each has to be examined by a doctor. This is very time consuming, demands heightened attention over a long time, which can easily cause erroneous diagnosis. Therefore computerized processing of the video is needed, that is capable of discovering symptoms of specific diseases and disorders without human intervention, thereby reducing the number of images to be examined manually by the doctor.
The subject of my thesis is bleeding detection in the gastrointestinal tract. At first I studied the relevant literature, and selected a well established algorithm. Then I created a suitable test set of images to verify the functioning of this algorithm, which is consists of real capsule endoscopy images, containing images of bleedings, and of normal tissues. Finally, I improved upon the results by determining the proper parameters.