A4 Conference proceedings

Weight Averaging Impact on the Uncertainty of Retinal Artery-Venous Segmentation

Open Access publication

Publication Details
Authors: Lindén Markus, Garifullin Azat, Lensu Lasse
Publisher: Springer Verlag (Germany): Series
Publication year: 2020
Language: English
Related Journal or Series Information: Lecture Notes in Computer Science
Title of parent publication: Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Graphs in Biomedical Image Analysis
Journal name in source: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Journal acronym: LNCS
Volume number: 12443
Start page: 52
End page: 60
Number of pages: 9
ISBN: 978-3-030-60364-9
eISBN: 978-3-030-60365-6
ISSN: 0302-9743
eISSN: 1611-3349
JUFO-Level of this publication: 1
Open Access: Open Access publication
Location of the parallel saved publication: http://urn.fi/URN:NBN:fi-fe2020120198841


By examining the vessel structure of the eye through retinal imaging, a variety of abnormalities can be identified. Owing to this, retinal images have an important role in the diagnosis of ocular diseases. The possibility of performing computer aided artery-vein segmentation has been the focus of several studies during the recent years and deep neural networks have become the most popular tool used in artery-vein segmentation. In this work, a Bayesian deep neural network is used for artery-vein segmentation. Two algorithms, that is, stochastic weight averaging and stochastic weight averaging Gaussian are studied to improve the performance of the neural network. The experiments, conducted on the RITE and DRIVE data sets, and results are provided along side uncertainty quantification analysis. Based on the experiments, weight averaging techniques improve the performance of the network.

KeywordsArtery-vein segmentation, Bayesian deep learning, Blood vessel segmentation, Weight averaging

Last updated on 2020-01-12 at 10:06