ELM for Retinal Vessel Classification
Abstract
Robust image segmentation can be achieved by pixel classification based on features extracted from the image. Retinal vessel quantification is an important component of retinal disease screening protocols. Some vessel parameters are potential biomarkers for the diagnosis of several diseases. Specifically, the arterio-venular ratio (AVR) has been proposed as a biomarker for Diabetic retinopathy and other diseases. Classification of retinal vessel pixels into arteries or veins is required for computing AVR. This paper compares Extreme Learning Machines (ELM) with other state-of-the-art classifier building approaches for this tasks, finding that ELM approaches improve over most of them in classification accuracy and computational time load. Experiments are performed on a well known benchmark dataset of retinal images.
BIB_text
author = {Iñigo Barandiaran, Odei Maiz, Ion Marqués, Jurgi Ugarte, Manuel Graña},
title = {ELM for Retinal Vessel Classification},
pages = {135-145},
volume = {16},
keywds = {
fundus, retina, ELM, classification
}
abstract = {
Robust image segmentation can be achieved by pixel classification based on features extracted from the image. Retinal vessel quantification is an important component of retinal disease screening protocols. Some vessel parameters are potential biomarkers for the diagnosis of several diseases. Specifically, the arterio-venular ratio (AVR) has been proposed as a biomarker for Diabetic retinopathy and other diseases. Classification of retinal vessel pixels into arteries or veins is required for computing AVR. This paper compares Extreme Learning Machines (ELM) with other state-of-the-art classifier building approaches for this tasks, finding that ELM approaches improve over most of them in classification accuracy and computational time load. Experiments are performed on a well known benchmark dataset of retinal images.
}
isbn = {978-3-319-04741-6},
date = {2013-10-15},
year = {2013},
}
