1. Title: Contractions database Two-class problem: detecting intestinal contractions in video images (capsule endoscopy) 2. Sources: (a) Creator: Fernando Vilarino Computer Vision Centre, Universitat Autonoma de Barcelona, 08193, Bellaterra Barcelona SPAIN (c) Date: December 2004 3. Past Usage: 1. F. Vilarino, L. I. Kuncheva and P. Radeva, ROC curves and Video Analysis Optimization in Intestinal Capsule Endoscopy, Pattern Recognition Letters, Special Issue on ROC Analysis, accepted. (http://www.informatics.bangor.ac.uk/~kuncheva/papers/fvlkprprl.pdf) 4. Relevant Information: Wireless capsule video endoscopy is a recent technology in which a pill with an attached camera is swallowed by the patient. The camera travels along the intestinal tract and emits a radio signal recorded as a video. The problem is to detection of contraction frames in the video automatically. Each contraction is defined as a sequence of 9 image frames. 27 attributes were derived from these frames, 3 features for each frame: mean intensity, the hole size (lumen opening) and global contrast. The 9 values from the consecutive frames have been standardized by taking out the mean and dividing by the standard deviation of the respective feature. 5. Number of Instances: 98 6. Number of Attributes: 27 (all continuous-valued) plus the class attribute 7. Attribute Information: 1. - 9. mean intensity of frames 1 to 9 10. - 18. hole size of frames 1 to 9 19. - 27. global contrast of frames 1 to 9 28. Class label (1 - contractions; 2 - non-contractions) 8. Missing Attribute Values: None 9. Class Distribution: (out of 98 total instances) -- 49 from Class 1 -- contractions -- 49 from Class 2 -- non-contractions