Elsevier

Information Sciences

Volume 441, May 2018, Pages 41-49
Information Sciences

Deep convolution neural network for accurate diagnosis of glaucoma using digital fundus images

https://doi.org/10.1016/j.ins.2018.01.051Get rights and content

Highlights

  • An eighteen layer CNN framework is proposed for glaucoma diagnosis.

  • Used 1426 fundus images (normal: 589 and glaucoma: 837).

  • Attained accuracy of 98.13%, sensitivity of 98% and specificity of 98.3%.

Abstract

Glaucoma progressively affects the optic nerve and may cause partial or complete vision loss. Raised intravascular pressure is the only factor which can be modified to prevent blindness from this condition. Accurate early detection and continuous screening may prevent the vision loss. Computer aided diagnosis (CAD) is a non-invasive technique which can detect the glaucoma in its early stage using digital fundus images. Developing such a system require diverse huge database in order to reach optimum performance. This paper proposes a novel CAD tool for the accurate detection of glaucoma using deep learning technique. An eighteen layer convolutional neural networks (CNN) is effectively trained in order to extract robust features from the digital fundus images. Finally these features are classified into normal and glaucoma classes during testing. We have achieved the highest accuracy of 98.13% using 1426 (589: normal and 837: glaucoma) fundus images. Our experimental results demonstrates the robustness of the system, which can be used as a supplementary tool for the clinicians to validate their decisions.

Introduction

Glaucoma refers to the progressive loss of the retinal ganglion cells (RGC) leading to the visual loss. It results in varying degrees of irreversible visual disability and in some cases blindness. According to the World Health Organization (WHO), glaucoma is the second leading cause of blindness worldwide after cataracts [1], [38]. The retinal nerve fibers are typically represented by the annular region between optic disc and the cup boundary, which is known as neuro retinal rim [2], [36]. The fluid pressure in the inner portion of the eye is called intraocular pressure (IOP) [35], [39]. Increase in this IOP leads to blockage of outflow of aqueous humor. This damages the optic nerve, which is essential to communicate the information from retina to the brain (Ref. Fig. 1) [37], [38]. This deterioration of optic nerve fibers results in thickening of retinal nerve fiber layer (RNFL), which is usually known as ‘cupping’. This cupping causes the progression of glaucoma [36], [40]. The decrease in the healthy neuro retinal tissues can be easily noticed by measuring cup to disc ratio (CDR), which is an indication of glaucomatous change [40]. Typical value of CDR for a healthy eye is 0.3 [39], [41].

The clinical diagnosis of glaucoma includes a series of various tests that is carried out by the ophthalmologist. The key to prevent glaucoma is to have regular eye check-ups after 40 years. The following different tests are commonly performed to diagnose glaucoma are described as follows [3]:

Tonometry measures the pressure within the eye. During tonometry, eye drops are used to numb the eye. Then a doctor or technician uses a device called a tonometer to measure the inner pressure of the eye. If it exceeds >21 mm Hg, the person is diagnosed with glaucoma [4]. In Ophthalmoscopy the doctor examine the optic nerve for glaucoma damage. Eye drops are used to dilate the pupil so that the doctor can see through the eye to examine the shape and color of the optic nerve. If the intraocular pressure is not within the normal range or if the optic nerve looks unusual, the patient has to go through Perimetry which is a test to chart the visual field. High IOP in the eye can damage OD in glaucoma. Gonioscopy [4]. Pachymetry is a simple test, which measures cornea thickness. It helps in clinical diagnosis as it has great influence on eye pressure reading [4].

The clinical diagnosis of eye using above techniques is time consuming and involves inter/intra observer variability. Digital fundus images captured using a fundus camera can be effectively utilized for observing the progression of the diabetic retinopathy (DR), glaucoma and age related macular degeneration (AMD). The interesting clinical features of eye such as retina, optic disc, blood vessels etc., can be clearly visualized in fundus images. In addition, fundus camera is reliable, less expensive and easy to operate and it can be used to measure various structures such as change in cup to disc ratio, optic nerve head (ONH), cup diameter etc. [41]. Hence, fundus images can be effectively utilized as a cost effective tool for the diagnosis of retinal health [35], [42] and eye abnormalities (DR, AMD and glaucoma) using a single fundus image.

Computer aided diagnosis (CAD) of fundus images helps to diagnose the retinal health using various computational algorithms. It is a cost effective tool which can avoid inter/intra observer variability which may be encountered in clinical diagnosis.

CAD has a major role in the diagnosis of glaucoma which can bring down the inter/intra observer variability. Also it requires minimum diagnosis time for more number of patients [5], [6], [18], [34], [35].

The detailed reviews of the existing work on CAD for glaucoma using digital fundus image are summarized in Table 1. It is observed that most of the algorithms follow the two stage pipeline structure: feature extraction and classification. The Wavelet Transform [9], [10], [14], [15], [22], Gabor transform [5], Higher order spectra (HOS) [8], [12], [23], etc., are the feature extraction techniques. In classification, artificial neural network (ANN) [2], [14], [21], support vector machine (SVM) [5], [9], [10], K-nearest neighbor (KNN) [13], [19], [24], etc., are used to predict the classes. The design of such hand-crafted features is tedious job and time consuming. These features are strongly related to expert knowledge and having restricted representation power. Thus, for a huge dataset it cannot show the discriminative power. To overcome this problem and enhance the classification performance, deep features are essential.

Although there have been a significant number of techniques proposed in the literature, it is required to develop an efficient algorithm using maximum number of subjects. Raghavendra et al. [45] have developed a novel expert system using GIST descriptor and SVM classifier. They achieved maximum performance of 97% accuracy for nineteen features using 1000 fundus images. Recently, Chen et al. [43] have used CNN for glaucoma detection by using region of interest (ROI) segmented images to select the optic nerve head from the images. But, our proposed work is fully automated deep learning architecture which can classify even early stage of glaucoma. Generally, convolution neural networks (CNN) extract the localized features from input images and convolution is performed with image patches using filters. Filter responses are pooled repeatedly and refiltered, and the output feature vector of resulting deep feed-forward network architecture are eventually classified [25], [26], [27]. The details of the developed model is presented in the subsequent sections.

Section snippets

Deep learning framework

Nowadays, image classification is performed using deep learning technique [29], [30], [31]. It integrates both feature extraction and classification. These methods can achieve promising results using complex networks built with large scale data. In this work, we have proposed a novel convolutional neural network (CNN) architecture to detect the glaucoma automatically with highest performance. During training our network, the features are extracted from the input images yielding a robust deep

Image dataset

The retinal fundus images have been obtained from Kasturba Medical College, Manipal, India. There are a total of 589 normal images and 837 glaucoma images. Zeiss FF 450 Fundus camera is used to capture all the images and all acquired images are converted to jpeg format for further processing. Institution ethical approval was taken before acquiring the fundus images. The sample fundus images are shown in Fig. 3.

Results

In order to perform the systematic evaluation, we have resized all the input fundus

Discussion

Generally, maximum number of images need to be used with CNN for better performance [7], [33]. In this work, we have used 1426 fundus images to develop the model which consists of eighteen layers as shown in Fig. 2 and Table 2. At learning rate 0.1, 0.01 and 0.0001, the results are comparatively low. This is because at this particular learning rate, the network may might miss the subtle important information. The model attained the best performance for 0.001 learning rate by gradually capturing

Conclusion

Glaucoma is one of the leading causes of blindness. A large number of populations across the globe is affected by it. The development of CAD tool will assist the ophthalmologists in detecting glaucoma even at mild stage. Our method obtained the highest accuracy, sensitivity, and specificity of 98.13%, 98%, and 98.30%, respectively, for the maximum number of images. The developed CNN model can efficiently detect the class (normal or glaucoma) of unknown image. The main advantage of CNN is, it

References (45)

  • K. Barton et al.

    Medical management of glaucoma

    Medical Management of Glaucoma

    (2013)
  • J. Nayak et al.

    Automated diagnosis of glaucoma using digital fundus images

    J. Med. Syst.

    (2009)
  • http://www.mayoclinic.org/diseases-conditions/glaucoma/basics/symptoms/con-20024042 (accessed 14 June,...
  • http://www.glaucoma.org/glaucoma/diagnostic-tests.php (accessed 21 July,...
  • U. Rajendra Acharya et al.

    Decision support system for the glaucoma using Gabor transformation

    Biomed. Signal Process. Control

    (2015)
  • U. Rajendra Acharya, H. Fujita, O. S. Lih, Y. Hagiwara, J. H. Tan, and M. Adam. ``Automated detection of arrhythmias...
  • S. Maheshwari et al.

    Automated diagnosis of glaucoma using empirical wavelet transform and correntropy features extracted from fundus images

    IEEE J. Biomed. Health Inf.

    (2016)
  • S. Dua et al.

    Wavelet-based energy features for glaucomatous image classification

    IEEE Trans. Inf. Technol. Biomed.

    (2012)
  • M.R.K. Mookiah et al.

    Automated glaucoma detection using hybrid feature extraction in retinal fundus images

    J. Mech. Med. Biol.

    (2013)
  • S. Simonthomas et al.

    Automated diagnosis of glaucoma using Haralick texture features

  • R. Gayathri et al.

    Automated glaucoma detection system based on wavelet energy features and ANN

  • G.O. Gajbhiye et al.

    Automatic classification of glaucomatous images using wavelet and moment feature

  • Cited by (363)

    View all citing articles on Scopus
    View full text