Open AccessOn optimization of active contour methods in colonoscopy image segmentation
Author(s) -
Raneem Ismail,
Szilvia Nagy
Publication year - 2022
Publication title -
image analysis and stereology
Language(s) - English
Resource type - Journals
eISSN - 1854-5165
pISSN - 1580-3139
DOI - 10.5566/ias.2604
Subject(s) - artificial intelligence , sørensen–dice coefficient , active contour model , segmentation , computer science , computer vision , hausdorff distance , pattern recognition (psychology) , preprocessor , subtraction , dice , similarity (geometry) , background subtraction , image segmentation , mathematics , image (mathematics) , pixel , statistics , arithmetic
As colonoscopy is the standard screening approach for colorectal polyps, and the first step of the correct classification and the efficient automatic diagnostics is the accurate detection and segmentation of the existing polyps, it is worth researching systematically, how colonoscopy databases are responding to two of the most influential variational segmentation methods, the geodesic and Chan–Vese active contour methods. Due to the quality variation of the colonoscopy databases, pre-processing steps are made. Then, 14 various filtered images are evaluated as different inputs for the active contour methods using the Sørensen–Dice Similarity Coefficient as a performance measurement metric. The effects of the initial mask shape and its size together with the number of iterations, contraction bias and smoothness factor were studied. The number of necessary iterations depended on the image resolution. In general, the Chan-Vese method showed more efficiency to match the actual contour of the polyp than the geodesic one with an initial mask possibly located within the polyp area. Preprocessing such as reflection removal, background subtraction and mean or median filtering improve the Sørensen–Dice coefficient can improve by 0.3.