Peer Reviewed

1

DOI

10.17077/omia.1022

Conference Location

Munich, Germany

Publication Date

October 2015

Abstract

We present a comprehensive and fully automatic glaucoma detection approach that uses machine learning techniques over multiple informatics domains, consisting of personal profile data, genetic data, and retinal image data. This approach, referred to as MKLclm, enriches the feature set of the multiple kernel learning (MKL) framework through the incorporation of classemes, which represent the outputs of multiple class-specific classifiers trained from the data of each informatics domain. We validate our MKLclm framework on a population- based dataset consisting of 2258 subjects, achieving an AUC of 94.9% ± 1.7% and a specificity of 88.5% ± 2.7% at 85% sensitivity, which is significantly better than the current clinical standard of care which uses intraocular pressure (IOP) for glaucoma detection. The experiments also demonstrate that MKLclm outperforms the standard SVM method using data from individual domains, as well as the traditional MKL method, showing that this deeper integration of data from different informatics domains can lead to significant gains in holistic glaucoma diagnosis and screening.

Rights

Copyright © 2015 Yanwu Xu, Lixin Duan, Damon Wing Kee Wong, Tien Yin Wong, and Jiang Liu

Included in

Ophthalmology Commons

Share

COinS
 
Oct 9th, 12:00 AM Oct 9th, 12:00 AM

Glaucoma Detection by Learning from Multiple Informatics Domains

Munich, Germany

We present a comprehensive and fully automatic glaucoma detection approach that uses machine learning techniques over multiple informatics domains, consisting of personal profile data, genetic data, and retinal image data. This approach, referred to as MKLclm, enriches the feature set of the multiple kernel learning (MKL) framework through the incorporation of classemes, which represent the outputs of multiple class-specific classifiers trained from the data of each informatics domain. We validate our MKLclm framework on a population- based dataset consisting of 2258 subjects, achieving an AUC of 94.9% ± 1.7% and a specificity of 88.5% ± 2.7% at 85% sensitivity, which is significantly better than the current clinical standard of care which uses intraocular pressure (IOP) for glaucoma detection. The experiments also demonstrate that MKLclm outperforms the standard SVM method using data from individual domains, as well as the traditional MKL method, showing that this deeper integration of data from different informatics domains can lead to significant gains in holistic glaucoma diagnosis and screening.