Glaucoma risk index: Automated glaucoma detection from color fundus images

https://doi.org/10.1016/j.media.2009.12.006Get rights and content

Abstract

Glaucoma as a neurodegeneration of the optic nerve is one of the most common causes of blindness. Because revitalization of the degenerated nerve fibers of the optic nerve is impossible early detection of the disease is essential. This can be supported by a robust and automated mass-screening. We propose a novel automated glaucoma detection system that operates on inexpensive to acquire and widely used digital color fundus images. After a glaucoma specific preprocessing, different generic feature types are compressed by an appearance-based dimension reduction technique. Subsequently, a probabilistic two-stage classification scheme combines these features types to extract the novel Glaucoma Risk Index (GRI) that shows a reasonable glaucoma detection performance. On a sample set of 575 fundus images a classification accuracy of 80% has been achieved in a 5-fold cross-validation setup. The GRI gains a competitive area under ROC (AUC) of 88% compared to the established topography-based glaucoma probability score of scanning laser tomography with AUC of 87%. The proposed color fundus image-based GRI achieves a competitive and reliable detection performance on a low-priced modality by the statistical analysis of entire images of the optic nerve head.

Introduction

Glaucoma is one of the most common causes of blindness with a mean prevalence of 2.4% for all ages and of 4.7% for ages above 75 years (Klein et al., 1992). The disease is characterized by the progressive degeneration of optic nerve fibers and astrocytes showing a distinct pathogenetic image of the optic nerve head.

Glaucoma leads to (i) structural changes of the optic nerve head (ONH) and the nerve fiber layer and (ii) a simultaneous functional failure of the visual field. The structural changes are manifested by a slowly diminishing neuroretinal rim indicating a degeneration of axons and astrocytes of the optic nerve (Fig. 1).

As lost capabilities of the optic nerve cannot be recovered, early detection and subsequent treatment is essential for affected patients to preserve their vision (Michelson et al., 2008). Commonly, glaucoma diagnosis is based on the patient’s medical history, intraocular pressure, visual field loss tests and the manual assessment of the ONH via ophthalmoscopy or stereo fundus imaging (Lin et al., 2007). To additionally objectify the glaucoma stage and its progression geometric parameters of the ONH are documented. These geometric parameters measure ONH structures that are changing in case of glaucoma disease: optic disk diameter, optic disk area, cup diameter, rim area, mean cup depth, etc.

This contribution provides a data-driven framework extracting a novel glaucoma parameter from fundus images. Contrary to the established detection techniques, it does not require accurate measurements of geometric ONH structures as it performs a statistical data mining technique on the image patterns themselves. The proposed methodology can be transferred to other domains and might be able to extract further parameters providing new insights to other ophthalmic questions.

Section snippets

Background

The glaucoma disease is characterized by the degeneration of optic nerve fibers and astrocytes that is often accompanied by an increased intraocular pressure. Due to the loss of nerve fibers the Retinal Nerve Fiber Layer (RNFL) thickness is decreasing. In the course of disease, the interconnection between the photoreceptors and the visual cortex is reduced. In the worst case, the visual information of the photoreceptors can no longer be transmitted to the brain and visual field loss up to

The concept of Eigenimages for glaucoma detection

Due to the high variability of the ONH appearance, the established determination of geometric ONH parameters utilized for glaucoma detection is difficult to automate.

We consider the described situation for automated glaucoma detection similar to that one stated by Turk and Pentland (1991) for early face detection methods. Before the 1990’s face detection systems characterized faces by a set of geometric parameters such as normalized distances or ratios between characteristic facial landmarks (

Image preprocessing

The proposed appearance-based approach analyzes the entire input image data to capture the glaucoma characteristics. To emphasize these desired characteristics in the input data, the variations not related to the glaucoma disease are excluded from the images in a preprocessing step. This includes variations due to image acquisition, such as inhomogeneous illumination or different optic nerve head locations, but also retinal structures not directly related to glaucoma, e.g. the vessel tree.

Due

Feature extraction

The performed image preprocessing emphasizes glaucomatous variations among the images and allows a generic and appearance-based feature extraction. The high-dimensional preprocessed images P are statistically compressed by PCA to gain compact, and meaningful features f. To capture complementary image information we propose three different generic image representations with different spatial and frequency resolution for feature extraction.

Classification

In the final classification step (Fig. 3), a glaucoma probability and the associated class label such as “glaucoma” or “not glaucoma” is computed from the three different feature types f that will be denoted as the Glaucoma Risk Index (GRI).

Evaluation

Based on the presented fully automated processing procedure illustrated in Fig. 3, we achieved a novel probabilistic index that we refer to as Glaucoma Risk Index (GRI). In order to quantify its ability in detecting glaucoma from color fundus images the performance of GRI is first characterized in more detail by some key figures and a reliability analysis. Furthermore, its performance is compared to (i) that of glaucoma experts and (ii) to medically relevant and well established glaucoma

Discussion

A reliable and competitive scheme for an automatic glaucoma detection was presented. The majority of the subjects were correctly classified as already shown. To give a better impression on the classification outcome, Fig. 9 shows examples of correctly classified and misclassified fundus images.

Fig. 9a–c illustrates correctly classified controls characterized by a typical small cup. In contrast, an expanded cup denotes the correctly classified images of glaucomatous eyes (Fig. 9d–f) which

Conclusion

Based on color fundus photos of the eye, the presented procedure for robust and reliable appearance-based feature extraction allows the automated quantification of the probability to suffer from glaucoma disease.

Due to glaucoma specific preprocessing and the appropriate combination of generic features, we are able to successfully apply the generic data-driven approach for this medical classification task. The proposed two-stage classification scheme helps to combine classifiers of different

Acknowledgements

This contribution was supported by the German National Science Foundation (DFG) in context of Collaborative Research Center 539 subproject A4 (SFB 539 A4), the German Academic Exchange Service (DAAD, Germany) and Hungarian Scholarship Board (MÖB, Hungary). The authors gratefully acknowledge funding of the Erlangen Graduate School in Advanced Optical Technologies (SAOT) by the DFG in the framework of the excellence initiative. R. Bock was supported by the DFG and SAOT. J. Meier is stipendiary of

References (59)

  • P. Sharma et al.

    Diagnostic tools for glaucoma detection and management

    Surv. Ophthalmol.

    (2008)
  • R. Varma et al.

    Expert agreement in evaluating the optic disc for glaucoma

    Ophthalmology

    (1992)
  • G. Wollstein et al.

    Identification of early glaucoma cases with the scanning laser ophthalmoscope

    Ophthalmology

    (1998)
  • J. Xu et al.

    Optic disk feature extraction via modified deformable model technique for glaucoma analysis

    Pattern Recognit.

    (2007)
  • M.D. Abràmoff et al.

    Automated segmentation of the optic disc from stereo color photographs using physiologically plausible features

    Invest. Ophthalmol. Vis. Sci.

    (2007)
  • B. Al-Diri et al.

    An active contour model for segmenting and measuring retinal vessels

    IEEE Trans. Med. Imag.

    (2009)
  • L.M. Alencar et al.

    Comparison of HRT-3 glaucoma probability score and subjective stereophotograph assessment for prediction of progression in glaucoma

    Invest. Ophthalmol. Vis. Sci.

    (2008)
  • Bertalmio, M., Sapiro, G., Caselles, V., Ballester, C., 2000. Image inpainting. In: Proceedings of the 27th Annual...
  • P. Betz et al.

    Biometric study of the disc cup in open-angle glaucoma

    Graefes Arch. Clin. Exp. Ophthalmol.

    (1982)
  • M. Blanco et al.

    Localization and extraction of the optic disc using the fuzzy circular Hough transform

    Lect. Notes Comput. Sci.

    (2006)
  • Bledsoe, W.W., 1966. The model method in facial recognition. Tech. rep., Panoramic Research Inc., Palo Alto, CA, Rep....
  • Bock, R., Meier, J., Michelson, G., Nyúl, L.G., Hornegger, J., 2007. Classifying glaucoma with image-based features...
  • A. Can et al.

    Rapid automated tracing and feature extraction from retinal fundus images using direct exploratory algorithms

    IEEE Trans. Inform. Technol. Biomed.

    (1999)
  • J.F. Canny

    A computational approach to edge detection

    IEEE Trans. Pattern Anal. Mach. Intell.

    (1986)
  • B.C. Chauhan et al.

    Technique for detecting serial topographic changes in the optic disc and peripapillary retina using scanning laser tomography

    Invest. Ophthalmol. Vis. Sci.

    (2000)
  • P.H. Chen et al.

    A tutorial on ν-support vector machines

    Appl. Stoch. Models Business Ind.

    (2005)
  • E. DeLong et al.

    Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach

    Biometrics

    (1988)
  • EL-Manzalawy, Y., Honavar, V., 2005. WLSVM: Integrating LibSVM into Weka Environment....
  • D.C. Fernández et al.

    Automated detection of retinal layer structures on optical coherence tomography images

    Opt. Express

    (2005)
  • Cited by (298)

    • Automatic detection of glaucoma via fundus imaging and artificial intelligence: A review

      2023, Survey of Ophthalmology
      Citation Excerpt :

      Glaucoma is a chronic progressive optic neuropathy in which changes in the structure of the optic nerve head (ONH) (Fig. 1A) and retinal nerve fiber layer (RNFL) are associated with visual defects. Structural changes are manifested by a slow, yet progressive, narrowing of the neuroretinal rim, indicating degeneration of retinal ganglion cells axons, and astrocytes of the optic nerve (13). To evaluate the narrowing of the neuroretinal rim (NRR) the clinician needs to identify the boundary contours of the cup and disc.

    • Distance-based confidence generation and aggregation of classifier for unstructured road detection

      2022, Journal of King Saud University - Computer and Information Sciences
    View all citing articles on Scopus
    View full text