• PDF / 4,965,483 Bytes
• 9 Pages / 439.363 x 666.131 pts Page_size
• 62 Downloads / 178 Views

DOWNLOAD

REPORT

Institute for Infocomm Research, Agency for Science, Technology and Research, Singapore 2 National University of Singapore, Singapore

Abstract. Glaucoma, Pathological Myopia (PM), and Age-related Macular Degeneration (AMD) are three leading ocular diseases worldwide. Visual features extracted from retinal fundus images have been increasingly used for detecting these three diseases. In this paper, we present a discriminative feature selection model based on multi-task learning, which imposes the exclusive group lasso regularization for competitive sparse feature selection and the graph Laplacian regularization to embed the correlations among multiple diseases. Moreover, this multi-task linear discriminative model is able to simultaneously select sparse features and detect multiple ocular diseases. Extensive experiments are conducted to validate the proposed framework on the SiMES dataset. From the Area Under Curve (AUC) results in multiple ocular diseases classification, our method is shown to outperform the state-of-the-art algorithms.

1

Introduction

Many of the leading causes of vision impairment and blindness worldwide are irreversible and cannot be cured [1]. Glaucoma, Pathological Myopia (PM), and Age-related Macular Degeneration (AMD) are three leading ocular diseases worldwide. Early detection of these ocular diseases utilizing effective visual features is highly needed [2][11]. With the advancement of retinal fundus imaging, several computer-aided diagnosis (CAD) methods and systems have been developed to automatically detect these three leading ocular diseases from retinal fundus images [6][4]. However, current work mainly focus on detecting Glaucoma, PM, and AMD individually. Classifying these three leading diseases simultaneously is still an open research direction. There are some correlations among these three leading ocular diseases. In recent decades, the problem of low vision and blindness in elderly people became a major and socially significant issue. The number of patients having age-related macular degeneration (AMD) in association with glaucoma is growing all over the world [8], which attaches great medical and social value to this multiple diseases diagnosis problem. Moreover, in a recent study, myopic eyes are less likely to have AMD and diabetic retinopathy (DR) but more likely to have nuclear cataracts and glaucoma [9]. c Springer International Publishing Switzerland 2015  N. Navab et al. (Eds.): MICCAI 2015, Part II, LNCS 9350, pp. 11–19, 2015. DOI: 10.1007/978-3-319-24571-3_2

12

X. Chen et al.

Fig. 1. System overview of our proposed discriminative feature selection scheme for multiple ocular diseases classification. Here “Sub1” stands for a sub-type disease of a leading ocular disease. As indicated within red dashed borders in the right of the figure, a sparse feature will be learned for each disease. For better viewing, please see the color pdf file.

In this paper, we adopt a Multi-task Learning (MTL) based method for discriminatively selecting sparse features, harmoniously integrating

Recommend Documents

Correntropy induced loss based sparse robust graph regularized extreme learning machine for cancer classification

163 26 4MB

DMRAE: discriminative manifold regularized auto-encoder for sparse and robust feature learning

143 52 615KB

Sparse Portfolio Selection via Bayesian Multiple Testing

208 19 2MB

Discriminative Feature Selection via Multiclass Variable Memory Markov Model

185 61 746KB

An information-theoretic graph-based approach for feature selection

185 43 793KB

Joint Self-expression with Adaptive Graph for Unsupervised Feature Selection

162 41 65MB

Feature Subset Selection Using Sparse Principal Component Analysis and Multiclass Fault Classification Using Selected Fe

172 40 87MB

Univariate Feature Selection Techniques for Classification of Epileptic EEG Signals

161 75 20MB

Feature Selection Using Sparse Twin Bounded Support Vector Machine

208 93 95MB

Feature and Sample Size Selection for Malware Classification Process

192 79 57MB

Feature Selection Algorithms for Plant Leaf Classification: A Survey

179 69 29MB

Feature selection for improved classification accuracy targeting riverine sand mapping

192 12 3MB