Félix Castro Alexander Gelbukh Miguel González Mendoza (Eds.)

The LNAI series reports state-of-the-art results in artificial intelligence research, development, and education, at a high level and in both printed and electronic form. Enjoying tight cooperation with the R&D community, with numerous individuals, as well as with prestigious organizations and societies, LNAI has grown into the most comprehensive artificial intelligence research forum available. The scope of LNAI spans the whole range of artificial intelligence and intelligent information processing including interdisciplinary topics in a variety of application fields. The type of material published traditionally includes ■ ■ ■

proceedings (published in time for the respective conference) post-proceedings (consisting of thoroughly revised final full papers) research monographs (which may be based on PhD work)

In parallel to the printed book, each new volume is published electronically in LNCS Online. Detailed information on LNCS can be found at www.springer.com/lncs Proposals for publication should be sent to LNCS Editorial, Tiergartenstr. 17, 69121 Heidelberg, Germany E-mail: [email protected]

1 LNAI 8265

Advances in Artificial Intelligence and Its Applications

More recently, several color-cover sublines have been added featuring, beyond a collection of papers, various added-value components; these sublines include ■ tutorials (textbook-like monographs or collections of lectures given at advanced courses) ■ state-of-the-art surveys (offering complete and mediated coverage of a topic) ■ hot topics (introducing emergent topics to the broader community)

Castro · Gelbukh Mendoza (Eds.)

Subseries of Lecture Notes in Computer Science

LNAI 8265

Lecture Notes in Artificial Intelligence

Advances in Artificial Intelligence and Its Applications 12th Mexican International Conference on Artificial Intelligence, MICAI 2013 Mexico City, Mexico, November 2013 Proceedings, Part I

ISSN 0302-9743 ISBN 978-3-642-45113-3

9 783642 451133

›

springer.com

123

Detection of Human Retina Images Suspect of Glaucoma through the Vascular Bundle Displacement in the Optic Disc José Abel de la Fuente-Arriaga1, Edgardo Manuel Felipe-Riverón2,∗, and Eduardo Garduño-Calderón3 1 Tecnológico de Estudios Superiores de Jocotitlán, Carretera Toluca-Atlacomulco Km. 44.8, C.P. 50700, Estado de México, Mexico 2 Centro de Investigación en Computación, Instituto Politécnico Nacional, Juan de Dios Bátiz s/n, C.P. 07738, D.F., Mexico 3 Centro Oftalmológico de Atlacomulco, Libramiento Jorge Jiménez Cantú 1208, C.P. 50450, Estado de México, Mexico [email protected], [email protected], [email protected]

Abstract. This work presents a methodology for detecting human retina images suspect of glaucoma based on the measurement of displacement of the vascular bundle caused by the growth of the excavation or cup. The results achieved are due to the relative increase in size of the cup or excavation that causes a displacement of the blood vessel bundle to the superior, inferior and nasal optic disc areas. The method consists of the segmentation of the optic disc contour and the vascular bundle located within it, and calculation of its displacement from its normal position using the chessboard metric. The method was successful in 62 images of a total of 67, achieving an accuracy of 93.02% of sensitivity and 91.66% of specificity in the pre-diagnosis. Keywords: Glaucoma detection, vascular bundle displacement, excavation detection, optic papilla segmentation, chessboard metric.

1

Introduction

Application of noninvasive techniques in automatic retina analysis is an important area in medicine [1]. The information achieved from the analysis of these digital images permits to decide about the existence of ocular diseases as the glaucoma [2]. The optic disc or optic papilla is the clearest area in images of the rear pole of the retina. In a normal papilla, the vascular network coming out from the choroids travels through the center of the nervous fibers that constitutes the optic nerve, which goes through a tube-like structure toward the brain. Glaucoma, an ocular asymptomatic neuropathy, is caused by an excessive intraocular pressure that creates an excavation (or cup) in the papilla that damages the optic nerve. This excavation produces a thickening of the wall of the papilla, which moves the cluster of veins and arteries (called ∗

Corresponding author.

F. Castro, A. Gelbukh, and M. González (Eds.): MICAI 2013, Part I, LNAI 8265, pp. 520–531, 2013. © Springer-Verlag Berlin Heidelberg 2013

Detection of Human Retina Images Suspect of Glaucoma

521

also vascular bundle), toward the nasal side of the affected eye. In time, the optic nerve is damaged, and if it is not medically treated in time, it causes first a progressive irreversible loss of peripheral vision and finally leads to blindness. Between a 12% and a 15% of the total world population is affected by various degrees of blindness due to glaucoma [2, 3] and in Mexico glaucoma represents the second cause of blindness [4]. Consequently, it is important to work in the search of new methods for effective early detection of glaucoma. The clinical procedures for the diagnosis of glaucoma are: (a) Analysis of the clinical history; (b) measurement of the intraocular pressure; (c) analysis of alterations in the optic disc; and (d) functional study of the visual field (Campimetry test). We use the term pre-diagnosis, to emphasize that the (final) diagnosis is done exclusively by the physician specialized in glaucoma. This work presents a new method for the pre-diagnosis of glaucoma based in morphological alterations detected within the optic disc (initial test for the disease identification). The approach is based on the close relationship found between the vascular bundle displacement and the excavation growth within the optic disc in the superior, inferior and nasal zones. It is possible using the proposed method to classify normal images (with a physiological excavation) and glaucoma suspect images, even when they are in its initial stages of development. The method was tested using 67 retina images of 43 patients (20 healthy and 23 suspect patients), which 24 are retinal images from 20 normal patients, and the rest of 43 images are formed by 21 isolated images and 11 pairs of images from 23 suspect patients. Images of the database used in this research were supplied by a physician specialized in glaucoma and used without performing any pre-processing before use. This group of images is a part of the private collection of retinal images that belongs to the Center for Computing Research of the National Polytechnic Institute, Mexico.

2

Background

A medical procedure used in the detection of glaucoma consists of evaluating the morphological alterations in the optic disc, visible with the help of instruments such as a slit lamp; high-power convex lenses [7]; optical photographs of the retina [5]; retinal confocal tomography [8, 9]; laser polarimetry tomography [8, 9], and optical coherence tomography [8, 10]. The performance and reproducibility of papillary measurements in glaucomatous retinas have been successfully evaluated and compared in numerous investigations [11]. The analysis of glaucoma begins with the detection and evaluation of some parts of the retina, mainly the optic disc (or optic papilla), the excavation and the blood vessels located within it (vascular bundle), since in function of its characteristics is possible to reach to important and valuable clinical findings. The main characteristics more frequently analyzed are the measurement of the Cup/Disc ratio and the reviewing of the neuro-retinal rim thickness, which in normal eyes is thicker in the inferior zone. The images used in this research are all retina fundus images, namely those that are captured through the eye pupil.

522

J.A. de la Fuente-Arriaga, E.M. Felipe-Riverón, and E. Garduño-Calderón

Essentially, even with a large variety of methods for the detection of the optic disc and cup, based primarily on thresholding methods as Otsu [6] and in the use the standard deviation [12, 14]; dynamics border methods as the circular Hough transform [15, 24] and the active contours (Snakes) [20, 22, 23, 24]; interpolation methods of borders as spline [16, 22, 23], clustering methods [18, 19]; and the combination of some of them, such as the OOCE rule which uses the Otsu thresholding method, the Opening morphologic operator (defined as the erosion followed by an dilation), the Closing morphologic operator (defined as the dilation followed by an erosion) and the External border method [13], the success of the segmentation process had difficulties caused mainly by the natural coverage of the blood vessels coming from the choroid, since their leaving from and spreading on the retina cover parts of the optic disc. To solve this difficulty authors has attempted to reduce the effect of the blood vessel by the use of the Black-Top-Hat morphologic operator (defined as the difference between the original image and its morphological closing) for the detection of the blood vessels [17], and the Opening and Closing operators for remove them [12, 24], However, these problems sometimes even precludes the exact contour detection of the optic disc and the possible excavations. Despite this, many reported results on glaucoma detection are based only on the Cup/Disc ratio [13], where if the diameter of the cup (or its equivalent in area) exceeds to 0.4 of the diameter (or area) of the optic papilla, then the eye probably becomes glaucomatous and cause invariably blindness in the patient, such as [14, 15] and [16]; they show accuracy between 60% to 94% of sensitivity and 82% to 94.7% of specificity in pre-diagnosis. Other works measure this same characteristics using pairs of stereo retinal images to find the disparity of the corresponding points of both images [20] achieving accuracy of 87% of sensitivity and 82% of specificity. However, even achieving results within the reported international ranges, this remains a subject of much controversy. Other authors, have reviewed other characteristics, combining the Cup/Disc ratio and the analysis of vessel bends or kinks, which are small vessels that when they come out from the excavation provide physiological validation for the boundary cup [22, 23], achieving accuracy between 18.6% to 81.3% of sensitivity and 45.5% to 81.8% of specificity, and the ISNT rule [12, 18] and [19], where, instead of measuring the neuroretinal rim thickness, they measure the proportion of the blood vessels in the inferior, superior, nasal and temporal disc zones, achieving accuracy between 97.6% to 100% of sensitivity and 80% to 99.2% of specificity in the pre-diagnosis.

3

Methodology

This work presents a new classification method of retina images with excavation suspect of glaucoma. The method relies only on the analysis of morphological alterations that can be detected within the optic disc, even when the excavation is in the initial stage of glaucoma. The method proposes a different way for the detection of the excavation, where it is not necessary to detect the exact contours of the disc and the excavation, commonly required by current method. The method is based in the analysis of the vascular bundle displacement within the optic disc due to the excavation, since we have observed a near correlation between the excavation growth and the blood vessels

Detection of Human Retina Images Suspect of Glaucoma

523

displacement in the inferior, superior and nasal optic disc zones. The proposed method consists of the following steps: 1. RGB image acquisition; 2. Segmentation of the optic disc region; 3. Detection of a reference point in the excavation or cup; 4. Detection of centroids of three zones of the vascular bundle; 5. Measurement of the distance between the reference point of the cup to the three centroids. Now we will describe in detail each procedure. 1. The proposed method was evaluated in 43 patients (20 healthy and 23 diseased), using 67 RGB retina images (24 normal and 43 suspect) of size 720 pixels of width and 576 pixels of height, in a JPG or BMP graphic format, which exhibit normal and suspect retinas with glaucoma at different stages of development. Images were acquired with a conventional eye fundus camera [5] and previously evaluated by an ophthalmologist specializing in glaucoma. Figure 1 shows an eye fundus image of our collection with the most important anatomic parts indicated: a. Optic disc; b. excavation or cup; c. vascular bundle; d. blood vessels; e. macula and fovea, whose excavation according to a specialist is classified in this case as suspect of glaucoma, due to the excessive size of the excavation with respect to the size of the optic disc.

Fig. 1. Optical color eye fundus image classified as suspect of glaucoma showing the following anatomic parts: a. optic disc; b. excavation or cup; c. vascular bundle; d. blood vessels; e. macula and fovea

2. The objective of the optic disc region segmentation is to obtain approximately the diameter of the disc to be used in the process of result normalization. We use the method developed in [24] for the optic disc segmentation. This method consists of six steps from which we occupy only the first five because we need only the segmentation of the optic disc area; we do not use the optic disc contour segmentation mentioned in the sixth step. The method in general find the region of interest (ROI), the elimination of the arterioles and venules and finally segments the optic disc with the use of morphological operators and the combination of the circular Hough transform and active contours. The method has an accuracy of 92.5% and is robust in images with low contrast, high levels of noise and edge discontinuities. The segmentation of the image shown in Fig.1 appears in Fig.2, where (a) shows the ROI containing the optic disc in RGB color model; (b) segmented optic disc area, and (c) the segmented area shown in (b) superimposed to the image shown in (a). 3. The detection of a reference point in the excavation begins with the automatic segmentation of the excavation by the method developed in [13], which uses the OOCE (Otsu, Opening, Closing, External border) rule. There, an accuracy of 92.52% was achieved. Then, a reference point farther toward the temporal region is selected

524

J.A. de la Fuente-Arriaga, E.M. Felipe-Riverón, and E. Garduño-Calderón

as shown in Fig.3 marked with a cross enclosed by a rectangle, and denoted by the capital letter A. This zone was selected because it is where the excavation border is farther from the center of the optic disc. We assumed that the displacement of the vessel bundle to the left is a consequence of the growth of the excavation to the right.

Fig. 2. Details of the optic disc contour which results from applying the algorithm detailed in [24]

Fig. 3. Reference point in the excavation located at the temporal optic disc zone

4. The detection of centroids of three zones of the vascular bundle (superior, inferior and nasal zones) is required for analyzing the displacement of blood vessels located within the optic disc. For this we take into account some reference points in the vascular bundle. These points approximately describe the tendency to grow in size of the excavation (located at the temporary zone) and the vascular bundle what is displaced to the nasal zone. This step begins with the segmentation of blood vessels located within the optic disc (the vascular bundle). For this, the central point of the optic disc is obtained in figure 2(c), which serves as a reference point for coupling it individually with the three triangular masks shown in Fig. 4. The central point of each mask must match the central point of the optic disc, with the purpose of the right choice of the target zone, even when the masks protrude from the ROI (the point does not always coincide with the center of the ROI). Figure 4(a) shows the mask for the superior zone; 4(b) for the inferior zone, and 4(c) for the nasal zone, in the case that the image to be analyzed is from the left eye. If the image to be analyzed is from the right eye, then the mask should be the mirror image of that shown in Fig. 4(c).

(a)

(b)

(c)

Fig. 4. Masks used for the segmentation of blood vessels located in different zones of the optic disc. (a) Mask for the superior zone; (b) for the inferior zone; (c) for the nasal zone (in case of the left eye).

Detection of Human Retina Images Suspect of Glaucoma

525

The blood vessel segmentation is carried out in the red channel of the RGB images, since the best results were achieved with this channel probably due to the common orange-reddish color of the optic papilla. The segmentation uses the morphologic Black-top-hat operator and the Otsu for thresholding method [17]. The Black-top-hat is a morphologic transformation defined as the difference between the original image and its morphological closing. This highlights the elements deleted by the closing which, in this case, are the blood vessels located within the optic disc. The result of the segmentation of blood vessels in the different zones is shown in Fig. 5, (a) superior zone; (b) inferior zone and (c) nasal zone.

(a)

(b)

(c)

Fig. 5. Segmentation of blood vessels located in the different zones of the optic disc, (a) superior zone; (b) inferior zone; (c) nasal zone

To find the reference point that describes the trend in the position of vessels being analyzed, we calculate the centroid for each zone. The centroid of a body coincides with the center of mass if the object density (area) is homogeneous or when the material distribution is symmetrical; however, if an area has irregular borders defined by relative complex mathematical expressions, the simplest method is to calculate the area of the objects (vessels) with help of the individual summations described in (1) [21]. n

n

n

i =1

i =1

i =1

A =  ΔAi Qx =  yiΔAi Qy =  xiΔAi x =

Qx Qy y= A A

(1)

Where ΔAi is the area of the ith element; n is the number of elements; yi is the coordinate y of the ith centroid of the element; xi is the coordinate x of the ith centroid of the element; x is the coordinate x of the centroid of the object; y is the coordinate y of the centroid of the object. The positions of centroids are shown in Fig. 6. They are marked by a cross enclosed by a rectangle, and denoted by capital letters. The letter B represents the centroid of vessels in the superior zone; the letter C the centroid of vessels in the nasal zone and the letter D represents the centroid of vessels in the inferior zone.

Fig. 6. Centroid of blood vessels located in the different zones of the optic disc, B: Superior zone; C: Nasal zone, and D: Inferior zone

526

J.A. de la Fuente-Arriaga, E.M. Felipe-Riverón, and E. Garduño-Calderón

5. In this stage the blood vessels displacement is calculated with respect to the excavation growth, which is carried out by using the reference point of the cup (A in Fig. 3) and the vessel centroids (Fig. 6) in the zones defined by the masks (Fig. 5). The measurement of the displacement consists of obtaining the distances d1, d2 and d3 from the centroids B, C and D to the reference point A located in the temporal part of the excavation, as shown in Fig. 7. Since normally the vessel displacement is in the horizontal direction, distances are measured using the chessboard metric defined as:

d c = max(| x2 − x1 |, | y 2 − y1 |)

(2)

Where (x1, y1) are the coordinates of the first point and (x2, y2) are those of the second.

Fig. 7. Distance d1, d2 and d3 between the centroids denoted by B, C and D and the reference point of the excavation A

The distances are normalized with respect to the horizontal diameter of the optic disc Dh. The normalized distance dn is:

dn =

100 d Dh

(3)

Where Dh is the horizontal diameter of the optic disc and d is the distance to normalize. The final result of the blood vessels displacement with respect to the proportional growth of the excavation is calculated using the average of the three normalized distances just measured. For pre-diagnosing if an optic disc is suspect to suffer glaucoma or not, we need to use a cut point to decide the question. Tests were carried out with 67 images. The cut point was selected at a normalized distance of 45 pixels obtained empirically from our experiments. If the average of distances calculated was less than the cut point, then the retina was pre-diagnosed as normal. On the other hand, if the average of distances was equal or greater than 45, the retina is pre-diagnosed as suspect to have glaucoma.

4

Discussion of Results

Figure 8 shows some images from the collection of 67 images of human retinas which were analyzed using the proposed method. The image population for the discussion of results was obtained in function of images with normal and suspect excavation.

Detection of Human Retina Images Suspect of Glaucoma

527

The factors taken in account to select the test images population were: Images with suspect excavation; images with generalized thinning of the neuroretinal rim (a); having wide excavation and ostensible scleral holes (b); images with normal excavation. Rule ISNT preserved the excavation and the normal output of the papilla vessels (c), (d). Figure 9 shows in detail the regions of interest (ROI) with the centroids and the reference points detected in different zones. The results achieved in these test images, were correct in all cases. The optic discs of the figures (a) and (b) have suspect excavations and the figures (c) and (d) have normal excavations.

(a)

(b)

(c)

(d)

Fig. 8. Some images used for the test of the proposed method

(a)

(b)

(c)

(d)

Fig. 9. Detail of the regions of interest (ROI) of the optic disc with the centroids and reference points detected in different zones: (a) and (b) images with suspect excavation; (c) and (d) images with normal excavation

The proposed method was tested with 67 retina images of real patients from which 24 were normal and 43 suspect to have glaucoma. From achieved results, 40 images had True Positives (TP), 22 images had True Negatives (TN), 2 images had False Positives (FP) and 3 images had False negatives (FN). To calculate the Sensitivity and Specificity, we used the following expressions based on the results above:

 TP   40  Sensitivit y =  100 =  100 = 93 .02 %  TP + FN   40 + 3 

 TN   22  Specificit y =  100 =  100 = 91 .66 %  TN + FP   22 + 2 

(4)

Then, it was achieved a Sensitivity of 93.02% and a Specificity of 91.66%, with an area under the curve of 0.923. In Table 1 are shown our results and some related to other state of the art solutions. The number of samples and the origin of the retina image databases in each case are listed below the table. Conversely, these ranges allow appreciating a confidence interval of the diverse method accuracies, demonstrating that the results achieved by the proposed method are within the range of those achieved by other state of the art solutions.

528

J.A. de la Fuente-Arriaga, E.M. Felipe-Riverón, and E. Garduño-Calderón

Table 1. Comparison of our results with those achieved by other state of the art solutions #

Type of analysis

Sensitivity

Specificity

1

Cup/Disc ratio [13]1, [14]2, [15]3, [16]4

60% to 94%

82% to 94.7%

2

Cup/Disc ratio used pairs of stereo retinal images [20]5

87%

82%

3

Cup/Disc ratio and vessel bends [22]6, [23]7

18.6% to 81.3%

45% to 81.8%

4

Cup/Disc ratio and ISNT rule [12]8, [18]9, [19]10

97.6% to 100%

80% to 99.2%

5

The proposed method11

93.02%

91.66%

1

107 images, from the Center for Computing Research of IPN database.

2

140 images, from the Singapore Eye Research Institute database.

3

90 images, from the Manchester Royal Eye Hospital database.

4

45 image, from the Gifu University Hospital database.

5

98 images pairs, from the Gifu University Hospital database.

6

138 images, from the Particularly Ophthalmology Clinic database.

7

27 images, from the Singapore Eye Research Institute database.

8

61 images, from the Kasturba Medical College database.

9

550 images, from an Aravind Eye Hospital database.

10

36 images, from an Aravind Eye Hospital database.

11

67 images, from the Center for Computing Research of IPN database.

Figure 10 shows two normal images that were classified as suspect of glaucoma. Fig. 10(a) has a Cup/Disc ratio less than 0.4 and meets the exigencies of the ISNT rule. Fig. 10(b) has a Cup/Disc ratio equal to 0.4 and also meets the exigencies of the ISNT rule. Then, both images have a pre-diagnosis of normal given by the specialist.

(a)

(b)

Fig. 10. Normal images classified as suspect of glaucoma

Detection of Human Retina Images Suspect of Glaucoma

529

Figure 11 shows two out of three retinal images that caused false negative results. They are suspect of glaucoma (diagnosis given by the specialist) and the system gave a result of healthy optic disc. Fig. 11(a) has an excavation with inferior polar notch, and a lightly asymmetric Cup/Disc ratio 0.4H (horizontal) and 0.5V (vertical). Fig. 11(b) presents a loss of nervous fibers (it is observed the channel without fibers, below of the superior temporal vascular bundle) at XI (eleven o’clock) that coincides with the superior notch of the excavation, having a vertical predominance. Figure 11(c) shows another situation, even when this appears as normal image belongs to a pair of image (in the work was used 11 pairs of images) of a patient with glaucoma; that is not a common situation. This occurs frequently when the glaucoma is caused by a traumatism, where only one optic disc is strongly affected and the other remains as normal. The fact of the high Cup/Disc ratio asymmetry is the reason by which the optic disc of that eye does not satisfy the ISNT rule to be considered as suspect of glaucoma.

(a)

(b)

(c)

Fig. 11. Images that caused false negatives with the proposed method

As it is shown in Table 1, the sensitivity and the specificity in pre-diagnosis of our proposed method is within the results achieved in the state of the art related solutions. Analyzing the results in detail, images where our method failed, it was observed that the expert pre-diagnosis are suspect because apart of the growth characteristics of the excavation, there is also the presence of nicks and localized loss optical fiber. On this basis, we can say that the effectiveness of the method presented is high.

5

Conclusion

This paper presents a method for classifying color images of human retinas suspect of having glaucoma on the basis of the displacement of the vascular bundle produced caused by the excavation growth in the optic disc. A close relationship was found between the excavation growth and the vascular bundle displacement in the superior, inferior and nasal optic disc zones. The proposed methodology achieved an accuracy of the 93.02% of sensitivity and 91.66% of specificity in the pre-diagnosis, even with excavations in early stages of development. This shows that the proposed approach is effective and can be used as a method for developing new ophthalmic medical systems suitable for computerized pre-diagnosis.

530

J.A. de la Fuente-Arriaga, E.M. Felipe-Riverón, and E. Garduño-Calderón

Acknowledgments. The authors of this paper wish to thank the Centro de Investigación en Computación (CIC), Mexico; Research and Postgraduate Secretary (SIP), Mexico, and Instituto Politécnico Nacional (IPN), Mexico, for their economic support.

References 1. Gili-Manzanaro, G., Carrasco, F.C., Martín-Rodrigo, J.C., Yangüela-Rodilla, J., AriasPuente, A.: Análisis digital de la papila con cámara de fondo de ojo convencional: estudio de variabilidad. Arch. Soc. Esp. Oftalmol. de Madrid 79(3) (2004) 2. Consejo de salubridad general: Aspectos Generales. Centro Nacional de Excelencia Tecnológica en Salud: Diagnóstico y Tratamiento del Paciente Adulto con Glaucoma de Ángulo Abierto, 10–26. Secretaría de Salud, México (2009) 3. Elolia, R., Stokes, J.: Monograph Series on Aging-related Diseases: XI. Glaucoma. Chronic Diseases in Canada 19(4) (1998) 4. Felipe-Riverón, E.M., Morales-Güitrón, S.L., Ortiz-Yánez, S.: Detección automática de la normalidad de las retinas humanas. Tesis de Maestría en Ciencias de la Computación CICIPN, México (2004) 5. Gili Manzanaro, G.: Fotografía de fondo de ojo con filtros. Boletín de a Soc Oftalmol de Madrid (44) (2004) 6. Otsu, N.: A threshold Selection Method from Gray-Level Histogram. IEEE Trans. Systems, Man and Cybernetics SMC-9(1), 62–66 (1976) 7. Ansari-Shahrezaei, S., Maar, N., Biowski, R., Stur, M.: Biomicroscopic measurement of the optic disc with a high-power positive lens. Invest Ophthalmol. Vis. Sci. 42(1) (2001) 8. Fernández-Argones, L., Piloto-Díaz, I., Coba-Peña, M., Pérez-Tamayo, B., DomínguezRandulfe, M., Trujillo-Fonseca, K.: Sistemas de análisis digital de imágenes en el glaucoma. Rev. Cubana Invest. Biomédica 28(3) (2009) 9. Mendoza-Santiesteban, C.E., Santiesteban-Freixas, R., González-García, A., GonzálezHess, L., Mesa-Hernández, E., Perdomo-Trujillo, Y., Alemañy-Rubio, E., Eguia, F., VidalCasalís, S.: La tomografía de coherencia óptica en el diagnóstico de enfermedades de la retina y el nervio óptico. Rev. Cubana Oftalmol. 18(2) (2005) 10. Duch-Tuesta, S., Buchacra-Castellano, O.: Tomografía de coherencia óptica en glaucoma. Infothea 43 (2006) 11. De la Fuente-Arriaga, J.A., Garduño-Calderón, E., Cuevas-de la Rosa, F.J.: Estado actual de las técnicas computacionales para la valoración del disco óptico en glaucoma. Memorias del Congreso Mexiquense CTS+I 1(1), 155–161 (2011) 12. Nayak, J., Acharya, R.: Automated Diagnosis of Glaucoma Using Digital Fundus Images. J. Med. Syst. 33(5) (2009) 13. Felipe Riverón, E.M., del Toro Céspedes, M.: Measurement of Parameters of the Optic Disk in Ophthalmoscopic Color Images of Human Retina. In: Sanfeliu, A., Martínez Trinidad, J.F., Carrasco Ochoa, J.A. (eds.) CIARP 2004. LNCS, vol. 3287, pp. 661–668. Springer, Heidelberg (2004) 14. Liu, J., Wong, D.W.K., Lim, J.H., Li, H., Tan, N.M., Wong, T.Y.: ARGALI: an automatic cup-to-disc ratio measurement system for glaucoma detection and AnalysIs framework. In: Proc. of SPIE, Medical Imaging 2009: Computer-Aided Diagnosis, vol. 7260 (2009) 15. Abdel-Ghafar, A., Morris, A., Ritchings, B., Wood, C.: Detection and Characterization of the Optic Disk in Glaucoma and Diabetic Retinopathy. UMIST 88 (2004)

Detection of Human Retina Images Suspect of Glaucoma

531

16. Hatanaka, Y., Noudo, A., Sawada, A., Hara, T., Yamamoto, T., Fujita, H.: Automated Measurement of Cup to Disc Ratio Based on Line Profile Analysis in Retinal Images. In: 33rd Annual International Conference of the IEEE EMBS, Boston (2011) 17. Muramatsu, C., Nakagawa, T., Sawada, A., Hatanaka, Y., Hara, T., Yamamoto, T., Fujita, H.: Automated segmentation of optic disc region on retinal fundus photographs: Comparison of contour modeling and pixel classification methods. Computer Methods and Programs in Biomedicine 101(1) (2010) 18. Kavitha, S., Duraiswamy, K.: An efficient decision support system for detection of glaucoma in fundus images using ANFIS. International Journal of Advances in Engineering & Technology 6(1), 226–240 (2012) 19. Narasimhan, K., Vijayarekha, K.: An efficient automated system for glaucoma detection using fundus image. Journal of Theoretical and Applied Information Technology 33(1) (2011) 20. Muramatsu, C., Nakagawa, T., Sawada, A., Hatanaka, Y., Yamamoto, T., Fujita, H.: Automated determination of cup-to-disc ratio for classification of glaucomatous and normal eyes on stereo retinal fundus images. Journal of Biomedical Optics 16(9) (2011) 21. Gere, J.M., Goodno, B.J.: Repaso de centroides y momentos de inercia. CervantesGonzález, S.R: Mecánica de materiales. Cengage Learning, 901–927 (2009) 22. Joshi, G.D., Sivaswamy, J., Krishnadas, S.R.: Optic disk and cup segmentation from monocular color retinal images for glaucoma assessment. IEEE Transactions on Medical Imaging 30(6), 1192–1205 (2011) 23. Liu, J., Wong, D.W.K., Lim, J.H., Li, H., Tan, N.M., Wong, T.Y.: Automated detection of kinks from blood vessels for optic cup segmentation in retinal images. In: Proc. of SPIE, Medical Imaging 2009: Computer-Aided Diagnosis, vol. 7260, pp. 72601J-1–72601J-8 (2009) 24. De la Fuente-Arriaga, J.A., Felipe-Riverón, E.M., Garduño-Calderón, E.: Segmentación del disco óptico en imágenes de retina mediante la transformada de Hough y los contornos activos. Research in Computer Science 58, 117–131 (2012)