I. INTRODUCTION:

Hypertension is a universal problem that affects around one billion people worldwide. Hypertensive retinopathy is the term used to define damage to the retina and retinal circulation due to high blood pressure. High blood pressure can cause damage to the retinal blood vessels. Most of the people will not experience any visual symptoms in the earlier stages, however some of them may account decreased vision or headaches^[1]. It may also lead to cardiovascular diseases and finally leads to death. Hence the timely diagnosis and treatment of the disease is vital.

A major aim of treatment is to prevent, limit the target organ damage by detecting it in earlier stages. Vessel diameter calculation and finding artery vein ratio (AVR) are few important measures for finding the presence of the Hypertensive Retinopathy disease using fundus images. In this paper, we present a model system for the automatic identification of hypertensive retinopathy using image processing.

1.1 Optic Disk Detection:

Accurate Optic disc centre and boundary is very important, to define the region of interest for the determination of artery to vein ratio. To define the region of interest to determine artery to vein accurate segmentation of optic disk is a necessary prerequisite. Boundary of optic disk is determined by modified hough transform [14] and on deformable model [15]. Foracchia et al. [16] modelled main retinal vessel by two parabolas, which have common vertex, the OD center. Hoover and Goldbaum [17] used fuzzy convergence technique to locate the OD position. The OD centre is determined by the intersection between finite lines. Mendonca etal [18] used vessel convergence in entropy vessel direction

1.2 Blood Vessel Detection:

Retinal vessel extraction, vessel features like length and width measurement, tortuosity measurement plays a crucial role to diagnose different pathologies like hypertensive retinopathy, cardiovascular problem. Fraz etal [19] classifies the blood vessel extraction techniques available in literature under five groups namely matched filtering algorithm [20], pattern recognition technique [21], morphological operation [22], tracking algorithm and model based approach.

1.3 Vessel Classification:

The techniques in literature can be grouped under two headings namely tracking based method and colour based method. Li etal [23] used central light reflex,to classify artery and vein. Grisan [24] etal used fuzzy clustering algorithm to classify artery and vein. Niemeijer et al. [25] classified artery and vein centreline pixel using a supervised method with a help of twelve features.

1.4 Vessel Selection for AVR Computation:

Vessel selection normally based on certain hypothesis proposed in literature. The following set of hypothesis is used as a selection. Vessel segments over a bifurcation or cross over are not taken. Thin vessels also not taken in to consideration, since it is difficult to classify thin vessel as artery/vein. False positive vessels detected in the back ground are not considered. By taking all the hypothesis into account in the range of interest i.e. twice the optic disk radius, vessels are chosen for AVR Computation.

1.5 AVR Computation:

Brinchmann-Hansen and Heier [26] proposed Half-Height Full Width algorithm(HHFW) to measure artery to vein ratio. Gregson [27] applied thresholding to extract vessel , followed by thinning to extract centreline pixels. Mosquera etal [28] proposed a semiautomatic method ART-VENA method to measure AVR. Kumar etal [29] used unsupervised linear discriminant analysis diameter measurement to measure AVR.

II. MATERIALS AND METHODS:

A. INSPIRE-AVR AND VICAVR DATABASE:

INSPIRE (Iowa Normative Set for Processing Images of the REtina) -AVR database [7] which contains 40 color images of the vessels and optic disc and an arterio-venous ratio reference standard by the University of Wisconsin, Madison, WI, USA. All the images are in JPEG format with a resolution of 2392X2048 pixels with 8-bits per pixel per color plane. This is the acknowledged reference standard and has been used in most important studies associating AVR in disease diagnosis.

The VICAVR database currently includes 58 images. The images acquired with a TopCon non-mydriatic camera NW-100 model and are optic disc centered with a resolution of 768x584. The database includes the caliber of the vessels measured at different radii from the optic disc as well as the vessel type (artery/vein) labelled by three experts [30].

Color fundus image with a resolution of 768*576 pixels were collected from Deepam eye hospital, Chennai. Twenty five normal images and seventy six abnormal images were collected , classification is done with a guidance of senior ophthalmologist.

Figure1. Original fundus image (Inspire-AVR, VICAVR)

B. METHODOLOGY:

Figure2. Overview of Methodology

The overview of all the steps implemented to identify the presence of hypertensive retinopathy is presented above.

REGION OF INTEREST MARKING:

Region of interest (ROI) marking [9] begins with the detection of optical disc, which is usually the most luminous region in a retinal image. Initially the RGB image is converted to green channel because the optic disc appears brighter in this channel. The local intensity variation method [2] was used to convert the green channel image into a black and white image using a threshold to locate the centre and radius of the optic disc. Sometimes a few images have more intensity in various areas other than optic disc; it may possibly result in a mismatch of the detected areas with the optic disc. This problem was solved using morphological operations. The optic disc is highlighted and the ROI is set twice the radius of optic disc as shown in the figure 4.

Figure4. Optic disc detection and ROI marking

VESSEL SEGMENTATION:

It is the initial step to illustrate how the blood vessels have been segmented [12] from the background tissues present in a retinal image. Local Entropy Thresholding [4] method has been used for segmentation of blood vessels. The local entropy method neglects the disproportion between foreground and background. As an improvising, optimal thresholding [5] maximizes local entropy based on the foreground and background ratio. So larger the local entropy implies a better balance ratio, thereby offering better preservation of blood vessels.

Figure3. Output for segmentation

C. CLASSIFICATION INTO ARTERIES AND VEINS:

Ophthalmologists differentiate arteries and veins on the basis of color and thickness [11] of the vessels. Due to acquisition process, images are non-uniformly illuminated and exhibit contrast variation . Image normalization proposed in [30] is used before classification. Generally, arteries exhibit a higher intensity level than the veins in the red channel of the RGB image. So, primarily convert the RGB image to red channel image and segment it into 3X3 individual parts. The mean intensity of all the pixels within this 3X3 segmented part are calculated and compared with individual components. The component with intensity greater than the mean intensity of all pixels is classified as artery and with lower value as veins. Using color components [3] arteries are marked with red color and veins with blue and the region of marking (ROI) is done using the centre and radius as explained in the above section.

Figure5. Classification output

D. CALCULATION OF AVR:

After classification, the vessel width [13] has to be measured to determine the arterio-venous ratio [6]. Within the region of interest the maximum intensity vessel width is measured using the formula

where r(start)=starting point of the vessel row,

r(end)=end pointof the vessel row,

c(start)=start point of the vessel column

c(end)= end point of the vessel column

is the width of the smaller branch arteriole/vein and

is the width of the larger branch arteriole/vein . To obtain precise values of the widths, Parr-Hubbar formulas [10] are used. The advantage of these formulas is they are more robust against variability in the number of vessels measured [8] and independent of image scale. They are defined as follows [2].

The central retinal artery equivalent (CRAE) is calculated as

The central retinal vein equivalent (CRVE) is calculated as

From the calculated values of CRAE and CRVE, the arterio-venous ratio is defined as

E. ACCURACY CALCULATION:

Comparing the calculated AVR value obtained from the above method with the available ground truth data given by the Ophthalmologist to find the accuracy of the method used. The ground truth data is a reference standard given by the University of Wisconsin, Madison, WI, USA [1]. The accuracy is calculated using these formulas.

Error = ((Calculated AVR-Database value) /Database Value) *100 ----------------(5)

Accuracy = (100 - |Error|) will give the percentage of accuracy of this method.

The results obtained using the above accuracy formulae for first five images of INSPIRE-AVR [7] and VICAVR are tabulated below.

TABLE I. CALCULATED AVR VALUES AND ACCURACY MEASUREMENT

IMAGE NO.	DATABASE AVR VALUE	CALCULATED AVR VALUE	ACCURACY
1	0.7	0.7080	98.85714
2	0.63	0.7434	82
3	0.7	0.7242	96.5428
4	0.65	0.7611	82.90769
5	0.78	0.7483	95.93589

TABLE 2: Calculated AVR value of VICAVR Database

IMAGE NO.	Ground Truth Value	CALCULATED AVR VALUE	ACCURACY
1	0.71	0.705	99.295
2	0.7	0.72	97.1428
3	0.7	0.7242	96.54285
4	0.69	0.71	97.101
5	0.75	0.7581	98.92

F. CREATION OF GRAPHICAL USER INTERFACE (GUI):

Graphical User Interface [9] is a front-end that automates the given task. Its main aim is to enhance the efficiency and ease of use for the underlying logical design of a stored program. The GUI is built as an executable program that allows user to load retinal images and performs the AVR measurements. Finally, the designed GUI will display the status of the retinal image as abnormal (HR diseased) if the AVR value is less than 0.6 else as normal, which will help the health professional for further diagnosis of the disease. The GUI output for a sample image from the INSPIRE –AVR database shows the status as normal as shown in figure6.

Figure6. GUI output to an input image

III. RESULTS

The above steps of the methodology are performed on 40 images of INSPIRE-AVR [7] database and the performance measures [9] for the entire database is done on the basis of sensitivity (SN), specificity (SP) and accuracy (Acc) which are defined as below.

As determined by the Gold standard , according to the methodology above used TP (true positive) is the event of counting the actual number of abnormal images which shows the status as abnormal, FN (false negative) is the event of counting the actual number of abnormal images which shows the status as normal, TN (true negative) is the event of counting the actual number of normal images which shows the status as normal, FP (false positive) is the event of counting the actual number of normal images which shows the status as abnormal. By performing the above methodologies the values obtained for INSPIRE-AVR database are as follows.

TP (True Positive) = 6, FN (False Negative) = 1, TN (True Negative) = 30, FP (False Positive) = 3. By substituting these values in the given formulae, the obtained performance measures are tabulated below.

The values obtained for VICAVR TP=16, TN=36, FP=2,FN=4.By substituting these values in the formula, the obtained performance

The values obtained for the images collected from hospital are as follows

True positive =74 ,true negative 24,false positive =1,false negative =2

TABLE II.PERFORMANCE MEASURES

Database	Sensitivity	Specificity	Accuracy
INSPIRE-AVR	0.8571	0.909	0.9
VICAVR	0.8	0.9473	0.896

IV. CONCLUSION

The fundus image analysis for the identification of hypertensive retinopathy is a latest development. It is achieved by performing vessel segmentation, region of interest marking, classification of vessels, artery-vein ratio (AVR) calculation which presents a preliminary step for the detection of hypertensive retinopathy. But all of these methods require the automation process to display the status of the patient, which will help the medical professionals in the decision support as well as in the mass screening of hypertensive retinopathy. Hence the required functionalities are integrated through a Graphical User Interface (GUI) which assists the medical professional to perform further treatment according to medical protocols. In this paper the work is carried out on a small set of images, hence it needs to be tested on a large database for robustness.

V. REFERENCES:

[1] Niemeijer M, Xu X, Dumitrescu A, Gupta P, van Ginneken B, Folk J, Abramoff M. “Automated Measurement of the Arteriolar-To-Venular Width Ratio in Digital Color Fundus Photographs”, IEEE Trans Med Imaging, 2011 Jun 16.

[2] Ortiz D, Cubides M, Suarez A, Zequera M, Quiroga J, Gomez J, Arroyo N, “Support System for the Preventive Diagnosis of Hypertensive Retinopathy”,International conference of the IEEE EMBS,pp.5649-5652,September 2010.

[3] Kevin Noronha, Navya K T and Prabhakar K Nayak. “Support System for the Automated Detection of Hypertensive Retinopathy using Fundus Images”. Special Issue on International Conference on Electronic Design and Signal Processing ICEDSP,vol1 ,pp.7-11, February 2012.

[4] T. Chanwimaluang and G. Fan, "An Efficient Blood Vessel Detection Algorithm for Retinal Images using Local Entropy Thresholding", in Proc. of the 2003 IEEE International Symposium on Circuits and Systems, Bangkok, Thailand, May 25-28, 2003.

[5] Manjunath Gudur, Nanda. S, “Automatic Retinal Vessel Segmentation using Modified Local entropy Thresholding Algorithm’’, International Conference on Electronics and Communication Engineering, 28th April, 2013.

[6] Lara Tramontan, Enrico Grisan, Alfredo Ruggeri, An improved system for the automatic estimation of the Arteriolar-to-Venular diameter Ratio (AVR) in retinal images, 30th Annual International IEEE EMBS Conference Vancouver, British Columbia, Canada, August 20-24, 2008.

[7] INSPIRE (Iowa normative set for processing images of the retina) AVR dataset from http://webeye.ophth.uiowa.edu

[8] Lara Tramontan, Enrico Grisan, Alfredo Ruggeri, An automatic system for the estimation of the generalized Arteriolar narrowing in retinal images, Proceedings of the 29th Annual International Conference of the IEEE EMBS Cité Internationale, Lyon, France, August 23-26, 2007.

[9] Georgios C. Manikis, Vangelis Sakkalis, Xenophon Zabulis, Polykarpos Karamaounas, Areti Triantafyllou, Stella Douma, Chrysanthos Zamboulis, Kostas Marias, An Image Analysis Framework for the Early Assessment of Hypertensive Retinopathy Signs, Proceedings of the 3rd International Conference on E-Health and Bioengineering – EHB, November 2011.

[10] L. D. Hubbard, R. J. Brothers, W. N. King, L. X. Clegg, R. Klein, L.S. Cooper, A. R. Sharrett, M. D. Davis, and J. Cai, “Methods for evaluation of retinal microvascular abnormalities associated with hypertension/sclerosis in the atherosclerosis risk in communities study,” Ophthalmology, vol. 106, no. 12, pp. 2269–2280, 1999.

[11] Andrea Zamperini, Andrea Giachetti, Emanuele Trucco, Khai Sing Chin,” Effective features for artery-vein classification in digital fundus images”, 25th IEEE International Symposium on Computer-Based Medical Systems, CBMS 2012, Rome, Italy,2012.

[12] M.M. Fraz, P. Remagnino, A. Hoppe, B. Uyyanonvara, A.R. Rudnicka, C.G. Owen, S.A. Barman, Blood vessel segmentation methodologies in retinal images– A survey, computer methods and programs in biomedicine I08 (2012),pg no.407-433.

[13] Y. Jiang, A. Bainbridge-Smith, A. B. Morris, ‘Blood Vessel Tracking in Retinal Images’, Proceedings of Image and Vision Computing New Zealand 2007, pp. 126–131,Hamilton, New Zealand, December 2007.

[14] Sekhar S., Al-Nuaimy W., and Nandi A.K. “Automated localisation of retinal optic disk using hough transform”. ISBI 2008. 5th IEEE International Symposium on Biomedical Imaging, 15(6) ; 1577-1580, 2008.

[15] A. Osareh, M. Mirmehdi, B. T. Thomas, and R. Markham “Classification and localisation of diabetic-related eye disease”. In Proceedings of the 7th European Conference on Computer Vision-Part IV, pages 502-516, 2002.

[16] M. Foracchia, E. Grisan, and A. Ruggeri “ Detection of optic disc in retinal images by means of a geometrical model of vessel structure”. IEEE Transactions on Medical Imaging, 23(19):1189-1195, 2004.

[17] Hoover A. and Goldbaum M “ Locating the optic nerve in a retinal image using the fuzzy convergence of the blood vessels”. IEEE Transactions on Medical Imaging, 22(8):951- 958, 2003.

[18] A. M. Mendonca, F. Cardoso, A. V. Sousa, and A. J. C. Campilho. “Automaticlocalization of the optic disc in retinal ages based on the entropy of vascular directions” In ICIAR (2), pages 424- 431, 2012.

[19] M. M. Fraz, P. Remagnino, A. Hoppe, B. Uyyanonvara, A. R. Rudnicka, C. G. Owen,and S. A. Barman “Blood vessel segmentation methodologies in retinal images – a Survey” Comput. Methods Prog. Biomed., 108(1):407-433, October 2012.

[20] S. Chaudhuri, S. Chatterjee, N. Katz, M. Nelson, and M. Goldbaum.” Detection of blood vessels in retinal images using two-dimensional matched filters” IEEE Transactions on Medical Imaging, 8(3):263-269, September 1989

[21] M. Niemeijer, J.J. Staal, B. van Ginneken, M. Loog, and M.D. Abramo_. Comparative study of retinal vessel segmentation methods on a new publicly available database. In J. Michael Fitzpatrick and M. Sonka, editors, SPIE Medical Imaging, volume 5370, pages 648-656. SPIE, SPIE, 2004.

[22] A.M. Mendonca and A. Campilho. Segmentation of retinal blood vessels by combining the detection of centerlines and morphological reconstruction. IEEE Transaction on Medical Imaging, 25(9):1200 -1213, Sept. 2006.

[23] H. Li, W. Hsu, M. L. Lee, and H. Wang. A piecewise Gaussian model for profiling and differentiating retinal vessel”. In ICIP03, volume 1, pages 1069-1072, 2003

[24] Enrico Grisan and Alfredo Ruggeri. “A divide et impera strategy for automatic classification of retinal vessels into arteries and veins”. Engineering in Medicine and Biology Society, 2003. Proceedings of the 25th Annual International Conference of the IEEE, 1:890-893, 2003.

[25] M. Niemeijer, B. van Ginneken, and M. D. Abramoff, “ Automatic classification of retinal vessels into arteries and veins” Medical Imaging 2009: Computer-Aided Diagnosis. Edited by Karssemeijer, Nico; Giger, Maryellen L.. Proceedings of the SPIE, 7260:72601F -72601F-8, 2009.

[26] Olaf Brinchmann-Hansen and Halvor Heier. “Theoretical relations between light streak characteristics and optical properties of retinal vessels”. Acta Phthalmologica, 64(S179):33-37, 1986.

[27] P.H. Gregson, Z. Shen, R.C. Scott, and V. Kozousek. Automated grading of venous beading. Computers and Biomedical Research, 28(4):291-304, 1995.

[28] A. Mosquera, R. Dosil, V. Lebor_an, F. Pardo, F. Gomez-Ulla, B. Hayik, A. Pose, and M. Rodrguez “ Art-vena: Retinal vascular caliber measurement” Pattern Recognition and Image Analysis. 1st Iberian Conference on Pattern Recognition and Image Analysis, 2003.

[29] Dinesh K. Kumar, Behzad Aliahmad, and Hao Hao. “Retinal vessel diameter measurement using unsupervised linear discriminant analysis”. ISRN Ophthalmology, 2012.

[30] M. Foracchia, E. Grisan, and A. Ruggeri, “Luminosity and contrast normalization in retinal images,” Med. Image Anal., vol. 9, no. 3, pp. 179–190, 2005

[30 ] http://www.varpa.es/vicavr.html

Received on 30.08.2015 Modified on 16.09.2015

Research J. Pharm. and Tech. 8(11): Nov., 2015; Page 1534-1539

DOI: 10.5958/0974-360X.2015.00274.7