Multi-Scale Directional Vesselness Stamping based Localization and Segmentation of Polyps from Video Capsule Endoscopy


Colonic polyps may be precursors to colon cancer and identifying malignant polyps is an important element of early detection. Computer aided diagnosis has potential for increasing diagnostic performance in malignant polyp detection and classification. In our work, we consider segmenting adenomas (malignant) and hyperplasias (benign) polyps using wireless capsule endoscopy (WCE) imagery based on texture distinctiveness via a multi-scale directional vesselness stamping (MDVS) method. Adenomas have a large number of vessel capillaries whereas hyperplasias usually have very few. We present our vascularization features based localization and segmentation results for polyps from WCE images.

Example localizations/segmentations (in RGB space):

Example results from ademona and hyperplasia WCE polyps ROIs taken from RapidAtlas® (Given Imaging, Isreal).

First row: Input ROI images. Second row: MDVS results from [1]





Vascularization in FICE:

(Coming soon)


[1] V. B. S. Prasath, R. Pelapur and K. Palaniappan. Multi-scale directional vesselness stamping based segmentation for polyps from wireless capsule endoscopy. 36th Annual International Conference EMBS (IEEE EMBS/EMBC 2014), Chicago, USA. Proc. IEEE. Late breaking research posters paper. Poster available at figshare: 10.6084/m9.figshare.1087896

[2] V. B. S. Prasath, H. Kawanaka. Vascularization features for polyp localization in capsule endoscopy. IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp. 1740-1742, 2015. Poster available at figshare: 10.6084/m9.figshare.1585847

[3] V. B. S. Prasath. Vascularization from Flexible Imaging Color Enhancement (FICE) for polyp localization. Journal of Medicine and Life, 10(2), pp. 147-149, Apr 2017.

[4] Colorectal polyp localization and segmentation in video capsule endoscopy with multiscale vascularization features. In preparation.

Back to Endoscopy Project Main page. Back to Research Page.