An efficient method based on watershed amd rule-based merging for segmentation of 3-D histo-pathological images

• Published in: Pattern recognition 2001-07, Vol.34 (7), p.1449-1458
• Main Authors: Umesh Adiga, P S, Chaudhuri, B B
• Format: Article
• Language: English
• ISSN: 0031-3203
• DOI: 10.1016/S0031-3203(00)00076-5

This paper deals with the segmentation of 3-D histo-pathological images. Here we have presented a region-based segmentation method involving watershed algorithm and the rule-based merging technique. We have implemented a new method similar to flooding process for circumventing the inability to automatically mark the regional minima in small isolated objects. The 3-D histo-pathological images for testing the algorithm are obtained using confocal microscope in the form of a stack of optical sections. Normally, result of a classical watershed algorithm on grey-scale textured images such as tissue images is over-segmentation. We have proposed a rule-based heuristic merging technique to reduce the over-segmentation of cells. The tiny fragments of the cells and their parents are identified based on some heuristic rules and are merged together. Rule-based merging gives more than 90% accurate segmentation when compared to simple classical watershed extended to 3-D. Results are shown on 3-D images of prostate cancer tissue specimen. copyright 2001 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.