Peer Reviewed

1

DOI

10.17077/omia.1015

Conference Location

Boston, MA, USA

Publication Date

September 2014

Abstract

Segmentation of hemorrhages helps in improving the efficiency of computer assisted image analysis of diseases like diabetic retinopathy and hypertensive retinopathy. Hemorrhages are blood leakages lying in close proximity to blood vessels, which makes their delineation from blood vessels challenging. We use multiresolution morphological processing with a view of achieving perceptual grouping of the hemorrhagic candidates occurring in variable shapes, sizes and textures. We propose a novel method of suppression of candidates lying on blood vessels while attaining a good segmentation of true hemorrhages including the ones attached to the vessels. Evaluated on 191 images having different degrees of pathological severity, our method achieved > 82% sensitivity at < 7 false positives per image (FPPI). We further observe that the sensitivity is higher for candidates with bigger sizes.

Rights

Copyright © 2014, Garima Gupta, Keerthi Ram, S. Kulasekaran, Niranjan Joshi, Mohanasankar Sivaprakasam and Rashmin Gandhi.

Included in

Ophthalmology Commons

Share

COinS
 
Sep 14th, 12:00 AM Sep 14th, 12:00 AM

Detection of retinal hemorrhages in the presence of blood vessels

Boston, MA, USA

Segmentation of hemorrhages helps in improving the efficiency of computer assisted image analysis of diseases like diabetic retinopathy and hypertensive retinopathy. Hemorrhages are blood leakages lying in close proximity to blood vessels, which makes their delineation from blood vessels challenging. We use multiresolution morphological processing with a view of achieving perceptual grouping of the hemorrhagic candidates occurring in variable shapes, sizes and textures. We propose a novel method of suppression of candidates lying on blood vessels while attaining a good segmentation of true hemorrhages including the ones attached to the vessels. Evaluated on 191 images having different degrees of pathological severity, our method achieved > 82% sensitivity at < 7 false positives per image (FPPI). We further observe that the sensitivity is higher for candidates with bigger sizes.