Elsevier

Pattern Recognition

Volume 40, Issue 7, July 2007, Pages 2063-2076
Pattern Recognition

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

https://doi.org/10.1016/j.patcog.2006.10.015Get rights and content

Abstract

A deformable-model based approach is presented in this paper for robust detection of optic disk and cup boundaries. Earlier work on disk boundary detection up to now could not effectively solve the problem of vessel occlusion. The method proposed here improves and extends the original snake, which is essentially a deforming-only technique, in two aspects: knowledge-based clustering and smoothing update. The contour deforms to the location with minimum energy, and then self-clusters into two groups, i.e., edge-point group and uncertain-point group, which are finally updated by the combination of both local and global information. The modifications enable the proposed algorithm to become more accurate and robust to blood vessel occlusions, noises, ill-defined edges and fuzzy contour shapes. The comparative results on the 100 testing images show that the proposed method achieves better success rate (94%) when compared to those obtained by GVF-snake (12%) and modified ASM (82%). The proposed method is extended to detect the cup boundary and then extract the disk parameters for clinical application, which is a relatively new task in fundus image processing. The resulted cup-to-disk (C/D) ratio shows good consistency and compatibility when compared with the results from Heidelberg Retina Tomograph (HRT) under clinical validation.

Introduction

Optic disk with bright-white region inside called pallor is one of main components on the fundus image as shown in Fig. 1(a). It is the entrance of the optic nerve and blood vessels to the retina. The 3-D shape of the optic disk is an important indicator of various ophthalmic pathologies. For an instance, the optic disk becomes bigger and deeper in the eyes with glaucoma. Clinicians quantify the cupping of the optic disk to evaluate the progression of glaucoma, where cup is defined at certain depth down from the disk edges. Cup and disk boundaries act as the references to quantitative measurements of the disk parameters, such as cup-to-disk (C/D) vertical ratio, C/D area ratio, etc., which are the important parameters for diagnosis. The existing ophthalmic instrument used to analyze the optic disk such as Optical Coherence Tomography (OCT) and Heidelberg Retina Tomograph (HRT) are based on scanning laser technique that can provide a colorless or pseudo-color 3-D visualization. The clinicians must manually place the disk boundary on the 3-D image as the reference, and then the cup boundary can be generated from the disk contour based on 3-D depth information. Since the image created from OCT or HRT is not true color image, the 2-D color fundus image is still referred to by most clinicians for the estimation of disk and cup boundaries. As an effective solution to this, a fully automated approach of cup and disk boundary detection is presented in this paper to provide essential disk parameters for clinical analysis and pathological monitoring.

The optic disk generally appears as a bright circular or elliptic region. The methods of optic disk boundary detection can be separated into two steps: optic disk localization and disk boundary detection. Correct localization of the optic disk may improve the accuracy of disk boundary extraction. Many existing approaches can be used to locate the optic disk with reasonable success. Sinthanayothin [1] located the position of the optic disk by finding the region with the highest local variation in the intensity. Tamur [2] and Pinz [3] applied Hough transform to obtain optic disk center and the outer circle of disk boundary. Recently, a principal component analysis (PCA) model based approach was used in Ref. [4], and template matching was used in Refs. [5], [6], [7]. Hoover [8] utilized the geometric relationship between the optic disk and main blood vessels to identify the disk location, similar approaches were introduced in Refs. [9], [10]. Correctly locating the optic disk is the first and essential step for optic disk segmentation. Subsequently the disk center is estimated and used to initialize the disk boundary.

Interference of blood vessels is one of the main difficulties to segment the optic disk reliably and accurately. This problem is very similar to other boundary detection and image segmentation problems in medical imaging area that still require robust solution. Currently, deformable models offer a reasonable approach for boundary detection and image segmentation which can be roughly classified into two categories: free-form deformable models, such as snakes, and parametrically deformable models, such as active shape models (ASMs). Mendels et al. [11] and Osareh et al. [6] extracted the optic disk boundary by GVF-snake algorithm, in which the blood vessel was first removed by morphology in the preprocessing step. Walter et al. [12] also used morphological filtering techniques to remove the blood vessels and then detected the optic disk boundary by means of shade-correlation operation and watershed transformation. Although the morphology preprocessing helps reduce the effect of blood vessels, it could not totally remove the effect. The resulted boundary was distorted in the regions with outgoing vessels. Li and Chutatape [13], [14] used a PCA method to locate the optic disk and an ASM to refine the disk boundary. Although this approach could indirectly handle blood vessel occlusion problem with moderate accuracy, by using shape models, the fuzzy shapes of optic disk due to various pathological changes were not easy to be represented by a number of shape models, which might reduce the accuracy of the result. Parametrically deformable models (ASM method) are suitable for use when more specific shape information is available and the detected object has relatively uniform shape with limited variation. However, in optic disk boundary detection, pathological changes may arbitrarily deform the shape of optic disk and also distort the course of blood vessels. Hence, deformable templates may not be able to sufficiently encode various shapes of optic disk from different pathological changes. Lowell et al. [7] segmented the optic disk by a contour deformation method based on a global elliptic model and a local deformable model with variable edge-strength dependent stiffness. However, the authors indicated that the performance to the images with variably pathological changes still needed to be further improved. A level set approach was introduced in Ref. [15], which can segment the objects with arbitrarily complex shapes. The advantage of this approach is its ability to evolve the model in the presence of sharp corners, cusps, shapes with pieces and holes, etc. Nevertheless, many of these problems are different from those encountered in the boundary detection of optic disk. How to remove the influence of blood vessels is still the main difficulty to the employment of the above-mentioned methods to the optic disk boundary detection.

Cup is the depressed area inside the optic disk, hence the 3-D depth is the primary feature of the cup boundary, for which the automated detection is a relatively new task and challenging work in fundus image processing. So far, very few researchers focused their work on cup boundary detection due to the fact that the 3-D image is not easily available. Results of disk and cup boundaries were shown in Ref. [16], however, the authors did not clearly describe how to obtain these boundaries.

Since 3-D images are not generally available, some definitions are provided to estimate the cup boundary on 2-D images. Previously, the clinicians used the pallor to estimate the cup boundary, while pallor is defined as the area of maximum color contrast inside the disk area. However, in many cases, there is no obvious pallor in the disk area. Even if the pallor is clear, its edge is just close to, but not the exact cup edge. Currently, the bending of small blood vessels at the cup edge is used as a clue to measure the cup boundary. Nevertheless, this method can only provide several points of cup boundary in the area where there are small blood vessels; for the area without small blood vessels, the cup boundary is not easy to be estimated. Alternatively, 3-D depth information is a relatively reliable feature to estimate the cup contour. There are several different cup definitions based on depths. In Ref. [17], cup is defined to be 50μm down from a reference plane which is estimated from the periphery of retinal surface, Another definition [18], [19], [20], [21], [22] locates cup at 150, 120μm, 50%, 13, etc., drop from the optic disk edge to the deepest point as shown in Fig. 1(b). All these cup definitions only depend on the depth information, because the 3-D image obtained in most of these literatures is colorless or only with pseudo-color.

In this paper, a novel approach for automated detection of cup and disk boundaries is proposed based on free-form deformable model technique (snake). The boundaries are extracted based on the combined information of smoothness, gradient, depth, etc. The algorithm extends the original snake technique further in two aspects to directly solve the problem of the influence of blood vessels without affecting the accuracy.

Section snippets

Optic disk boundary detection

The original deforming-only snake process is modified and implemented with further two extensions. Firstly, after each deformation, the contour points are classified into edge-point cluster or uncertain-point cluster by knowledge-based unsupervised learning. Secondly, the contour is updated through variable sample numbers. The updating is self-adjusted using both global and local information so that the balance on contour stability and accuracy can be achieved. The proposed method of optic disk

Cup boundary detection

The available 3-D optic disk image and disk boundary are the preconditions to estimate the cup boundary. Then the clinical disk parameters could be obtained from the detected disk and cup boundaries for analysis, diagnosis, and monitoring. Fig. 6 gives the basic steps of disk parameter estimation. Compared with the disk boundary detection on the 2-D fundus image, the variable shapes and colors of cup make it more difficult to be automatically segmented. Moreover, in many images, the cup contour

Results of disk boundary detection and evaluation

One hundred fundus images provided by the National University Hospital were tested by the proposed modified deformable model method, where the weight ratio was set to be β1:β2:β3:β4=1:1.2:1:1. The results were then compared with GVF-snake [6] and modified ASM algorithm [13], [14] based on the same initial boundary. In the modified ASM algorithm, PCA based method for disk center localization was also used, but it did not improve the accuracy of final disk boundary compared with the result from

Conclusion

Reliable and fully automated extraction of optic disk parameters can be a valuable diagnostic-assisting resource for clinicians. Much of prior work has focused on optic disk boundary detection, however the vessel occlusion problem has not been well solved. The present work has made a few contributions by proposing a novel approach to disk boundary detection which directly solves the problem of blood vessel occlusions. The method is then extended to detect the cup boundary and estimate the

Summary

The 3-D shape of the optic disk is an important indicator of various ophthalmic pathologies. Clinicians quantify the cupping of the optic disk to evaluate the progression of the eye disease, where cup is defined at certain depth down from the disk edges. Cup and disk boundaries act as the references to quantitative measurements of the disk parameters, such as cup-to-disk (C/D) vertical ratio, etc., which are the important parameters for diagnosis. Reliable and fully automated extraction of

About the Author—JUAN XU received her Bachelor degree in Electronics Department from Peking University, Beijing, P.R. China in 2001, and the Ph.D. degree from School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore in 2006. She was a Software Engineer in the Stratech Systems, Singapore, from 2005 to 2006. Currently, she is a Research Scholar in the Department of Ophthalmology, School of Medicine, University of Pittsurgh, USA. Her research interests include

References (25)

  • H. Li et al.

    Boundary detection of optic disk by a modified ASM method

    Pattern Recognition

    (2003)
  • G.J. Jaffe et al.

    Optical coherence tomography to detect and manage retinal disease and glaucoma

    Am. J. Ophthalmol.

    (2004)
  • J. Xu et al.

    Auto-adjusted 3-D optic disk viewing from low-resolution stereo fundus image

    Int. J. Comput. Biol. Med.

    (2006)
  • C. Sinthanayothin et al.

    Automated localisation of the optic disc, fovea, and retinal blood vessels from digital colour fundus images

    British J. Ophthalmol.

    (1999)
  • S. Tamur et al.

    Zero-crossing interval correction in tracing eye-fundus blood vessels

    Pattern Recognition

    (1988)
  • A. Pinz et al.

    Mapping the human retina

    IEEE Trans. Med. Imaging

    (1998)
  • H. Li, O. Chutatape, Automatic location of optic disk in retinal images, in: Proceedings of the International...
  • M. Lalonde et al.

    Fast and robust optic disc detection using pyramidal decomposition and Hausdorff-based template matching

    IEEE Trans. Med. Imaging

    (2001)
  • A. Osareh, M. Mirmehd, B. Thomas, R. Markham, Comparison of colour spaces for optic disc localisation in retinal...
  • J. Lowell et al.

    Optic nerve head segmentation

    IEEE Trans. Med. Imaging

    (2004)
  • A. Hoover et al.

    Locating the optic nerve in a retinal image using the fuzzy convergence of the blood vessels

    IEEE Trans. Med. Imaging

    (2003)
  • M. Foracchia et al.

    Detection of optic disc in retinal images by means of a geometrical model of vessel structure

    IEEE Trans. Med. Imaging

    (2004)
  • Cited by (0)

    About the Author—JUAN XU received her Bachelor degree in Electronics Department from Peking University, Beijing, P.R. China in 2001, and the Ph.D. degree from School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore in 2006. She was a Software Engineer in the Stratech Systems, Singapore, from 2005 to 2006. Currently, she is a Research Scholar in the Department of Ophthalmology, School of Medicine, University of Pittsurgh, USA. Her research interests include Medical Image Processing, Computer Vision, Camera Calibration and Pattern Recognition.

    View full text