Circular morphological feature extraction for nuclear medicine facilitation

Nuclear medicine applications of computing have largely been on demand due to expensive and time consuming pathological processes for detection of diseases like cancer. Automation tools exploiting morphological characteristics have been employed successfully. Limitations of the existing automated tools in handling specific situations keep the domain alive for research aimed at effective solutions. Detection and counting of circular regions representing the blood cells or bacterial nuclei is an important task in such solutions. The paper investigates traditional image processing methods like region-thresholding, circular Hough transform, Gabor Annulus and Otsu´s discriminant in handling of overlapping regions. The paper contributes a semi-supervised method for circular feature extraction and automatic counting. The method is shown to have an edge over existing methods.