South Asian Journal ofEngineering and Technology Vol.7, No.1 (2018) 63–71

ISSN No: 2454-9614

INVESTIGATION ON DIABETIC RETINOPATHY IMAGES FOR LEAKAGE DETECTION BY THE EXTRACTION OF INTENSITY AND COMPACTNESS FEATURE USING SALIENCY MAP

R.Jayanthia, Dr.K.Bommannarajaa*, A.V.Mohanapriyaa#

a-Associate Professor, a* –Principal, a#-PG student

a, a# Nandha College of Technology, a* KPR Institute of Engineering & Technology

, ,

Abstract

Diagnosis of this disease in its earlier phase is very important. The detection and assessment of leakage in retinal fluorescein angiogram images is important for the management of a wide range of retinal diseases. A new, unsupervised technique to detect and quantify leakage in FA images is proposed in this work. A novel efficient way to enhance leakage regions by using the concept of saliency is implemented. Saliency indicates the relative importance of visual features. A new technique to generate multiscale saliency maps with integration of the intensity and compactness cues of super pixels for this specific application. Inspired by the fact that human vision is usually more concerned with objects than with individual pixels and the objects of interest may vary in size, at different levels to represent the given images, and the powerful simple linear iterative clustering (SLIC) method is employed.

Keywords:DiabeticRetinopathy, Saliency Map

Received: 7/12/2017, Revised: 3/2/2018 and Accepted: 9/4/2018

1. Introduction

A.FA Images

Fluorescein angiography (FA) is a type of imaging commonly used in ophthalmology clinics that provides a map of retinal vascular structure and function by highlighting blockage of, and leakage from, retinal vessels1[1]. The value of FA in differential diagnosis of retinal diseases, such as age-related macular degeneration (AMD) and diabetic retinopathy (DR), is well recognized. The technology was first described in 1961 and was introduced into mainstream ophthalmology by Gass in 1967.FA is useful in evaluating diabetic eye disease as it is currently the gold standard for evaluating the retinal vasculature, the part of the retina most affected by diabetes.

B. Interpreting Fluorescein Angiography in Diabetics

In patients with diabetic retinopathy, imaging photos from FA can show microaneurysms, which manifest as punctuate areas of hyper fluorescence. Patchy areas of hypo fluorescence can signify ischemia from nonperfused retinal capillaries [17]. Increases in the foveal a vascular zone from macular ischemia can be seen using FA, which may help explain vision loss in some diabetic patients. FA can also show abnormal blood vessels in the eye such as intraretinal microvascular abnormalities (IRMA) or retinal neovascularization. Since Fluorescein is partially unbound in the blood stream, it can leak out of incompetent blood vessels. The visualization of leakage of Fluorescein dye over time is useful in showing the breakdown of the blood-retinal barrier. This is best exemplified in diabetic macular edema, which is visualized as Fluorescein leakage over time in the macula. Retinal neovascularization also can cause Fluorescein leakage, and FA is a useful test to confirm the diagnosis of neovascularization of the disc and elsewhere in proliferative diabetic retinopathy

Fig.1. Fluorescein Angiography of the Eye

C. Wide field Fluorescein Angiography

More recently, widefield FA has been developed which has allowed for improved imaging of the peripheral retina Figure 2. This technology can be helpful in detecting the peripheral neovascularization, as well as the extent of retinal nonperfusion. Wide field FA may reveal peripheral areas of capillary nonperfusion that are difficult to visualize with standard field FA. It has been hypothesized that these nonperfused regions may be a source of vascular endothelial growth factor (VEGF), which might contribute to formation of diabetic macular edema. VEGF release might theoretically be halted by “targeted” panretinal photocoagulation (PRP) to these areas of photocoagulation (PRP) to these areas of C. Wide field Fluorescein Angiography

More recently, widefield FA has been developed which has allowed for improved imaging of the peripheral retina Figure 2. This technology can be helpful in detecting the peripheral neovascularization, as well as the extent of retinal nonperfusion. Wide field FA may reveal peripheral areas of capillary nonperfusion that are difficult to visualize with standard field FA. It has been hypothesized that these nonperfused regions may be a source of vascular endothelial growth factor (VEGF), which might contribute to formation of diabetic macular edema. VEGF release might theoretically be halted by “targeted” panretinal photocoagulation (PRP) to these areas of

Fig 2. Improved imaging of the peripheral retina

Wide field Fluorescein angiography in a patient with proliferative diabetic retinopathy. Note the numerous areas of leakage by the disc and along the arcades corresponding to neovascularization of the disc and elsewhere, respectively.

D. Limitations of FA

It is important to note that FA was not part of the ETDRS criteria for determining if a patient has CSME. Rather, CSME is an examination finding diagnosed by fundus biomicroscopy with one of three criteria: Hard exudates within 500 μ of the fovea with associated retinal thickening, retinal thickening within 500 μ of the fovea, or an area of retina thickening over 1500 μ in diameter that is less that 1500 μ from the fovea. Also unclear is the utility of FA for determining treatment parameters for macular edema or determining success of macular edema therapy [12].In addition, visualization of deep retinal and choroid vessels is limited using FA. Fluorescein angiography also has a number of potential side effects.

The most common complications are transient nausea that occurs in about 2.9% of patients, as well as vomiting in 1.2% of patients [13].There are also various allergic reactions ranging from mild pruritus and urticaria to severe anaphylaxis and death, although the latter are extremely rare. There are no risks or adverse reactions that have been associated with pregnancy, but most physicians avoid performing this test in pregnant patients. It is of note that Fluorescein dye is not iodine-based, and patients with an iodine allergy can still receive Fluorescein dye. The percentage of reactions in patients who had had the previous FA without complication was 1.8%, compared with 48.6% in patients who had had a prior reaction to FA.

E. Leakages in Diabetic Retinopathy

This lining is called the retina. A healthy retina is necessary for good eyesight. Fluorescein angiography (FA) is a type of imaging commonly used in ophthalmology clinics that provides a map of retinal vascular structure and function by highlighting blockage of, and leakage from, retinal vessels[1].The value of FA in differential diagnosis of retinal diseases, such as age-related macular degeneration (AMD) and diabetic retinopathy (DR), is well recognized. Malarial retinopathy (DR) is made up of a collection of important signs in pediatric cerebral malaria (CM). DR reveals almost all the different types of angiographic vascular abnormalities that are common to other retinal conditions, and as such it is convenient to use this condition to develop semi-automatic or fully automatic tools for quantitatively assessing the retinal abnormalities.

On the other hand, similarities between eye and brain relevant to cerebral malaria seems to suggest that the retina may be a good source of potential biomarkers which might cast light on cerebral malaria disease processes [2].However, with few exceptions [3][4]existing descriptions of pediatric DR are qualitative or semi-quantitative, and based on ophthalmoscopic examination rather than more sophisticated imaging techniques. FA captures a range of retinal abnormalities in pediatric DR, in addition to those visible with ophthalmoscope. These include capillary non-perfusion3, intravascular filling defects, and several types of leakage. Retinal vessel leakage is particularly relevant to cerebral malaria, since the blood-retinal barrier is similar to the blood-brain barrier, and leakage from the latter could contribute to the brain swelling commonly seen in pediatric cerebral malaria.

2. EXISTING SYSTEM

A novel unsupervised technique to quantify the type and severity of leakage in MR by a novel adaptation of the concept of salient features. When one sees the world, the brain relies onattentionto capture the most salient details.Selective attentionrefers to the cognitive mechanism that determines which part of the plethora of sensory data is currently of most interest (i.e.salient) the cocktail party effectis a well-known example of selective attention. Visual saliency is of great importance in neurophysiology, psychology and computer vision. It indicates the relative importance of visual features and is closely related to characteristics of human perception and processing of visual stimuli17, 18. Such attributes are characteristic of retinal leakage in FA images. Figure 3 gives the work flow for the existing method used to find the leakage detection of Diabetic retinal images using simple linear iterative clustering (SLIC), which causes a large difference in brightness between the leak and surrounding non-leaking areas. Therefore, the leaking regions may be defined as salient regions. We now describe our saliency computation, which is based on this rule: the salient region is always different from is surrounding context.

The performance of this automated method against a human expert reference standard. By accurately quantifying neurovascular leakage, our method will facilitate analyses of associations between retinal vessel leakage and clinical outcome in cerebral malaria. It has similar potential for other systemic and ophthalmological conditions characterized by retinal leakage.

Fig.3 Work flow of Existing Method

3. PROPOSED WORK

A novel efficient way to enhance leakage regions by using the concept of saliency. Saliency indicates the relative importance of visual features, and is closely related to the characteristics of human perception and processing of visual stimuli. Saliency emerges from such characteristics in features of the image as visual uniqueness. Unpredictability, or rarity, and is often attributed to variations in specific image attributes such as color, gradient, edges, and boundaries. Such attributes are also characteristics of retinal leakage in FA images. A new way to generate multiscale saliency maps with integration of the intensity and compactness cues of super pixels for this specific application.

Figure 4 gives the work flow for the proposed method used to find the leakage detection of Diabetic retinal images using Fuzzy mean clustering simple linear iterative clustering (FCM SLIC), which gives better performance than the existing method. More specifically, traditional saliency extraction methods usually compute the salience of an image in a pixel-by-pixel manner, and ignore the neighbourhood and edge information of the objects of interest. Inspired by the fact that human vision is usually more concerned with objects than with individual pixels and the objects of interest may vary in size, it is proposed to use patches at different levels to represent the given images, and the powerful fuzzy based simple linear iterative clustering (SLIC) method is employed for this task.

Fig.4 Work flow of Proposed Method

3.B. Saliency Detection

The saliency saliency is determined as the local contrast of an image region with respect to its neighborhood at various scales.

4. A. Intensity Based Saliency Detection

Super pixel Pi is considered salient when it is highly dissimilar to other super pixels. The saliency value of Pi is defined as

4.B. Compactness Based Saliency Detection

Normally, human observers pay more attention to a more compact object than to a more diffuse object. The measure of compactness of an object might therefore be of use as a complementary feature to intensity for saliency measurement, with the aim of reducing the number of falsely-detected salient regions.

5. Graph Cut Segmentation

E (f) = Esmooth (f) + Edata (f)

Here Esmooth (f) measures the extent to which f is not piecewise smooth, while Edata (f) measures the disagreement between f and the observed data. The form of Edata (f) is typically,

Edata(f) = X p∈P Dp (fp)

Esmooth (f) = X {p,q}∈N u{p,q} .T(fp 6= fq)

Where T is indicator function i.e. The main problem, however, is that interesting energies are often difficult to minimize. E.g. it is proved that minimization of Potts energy on Grid Graph with multiple labels is NP hard. So, to find approximate solution, Boykov et. A proposed two algorithms named α−β swap and α-expansion. Also, using these moves solutions do not change significantly by varying the initial labelling

6.RESULTS AND DISCUSSIONS

TABLE 1.Result Using DR Datasets

S.NO / FEATURES / PERFORMANCE
1 / Sensitivity(Se) / 0.96±0.03
2 / Specificity(Sp) / 0.97±0.02
3 / Accuracy(Acc) / 0.93±0.03
4 / Ares under the receiver operating characteristic curve(AUC) / 0.97±0.02

TABLE 2 Parameters Comparisons

S.NO / FEATURES / FCM SLIC / SLIC
1 / Sensitivity(Se) / 97.2622 / 84.8495
2 / Specificity(Sp) / 97.9094 / 90.1482
3 / Accuracy(Acc) / 97.9027 / 89.9198
4 / Ares under the receiver operating characteristic curve(AUC) / 97.5836 / 87.4989

In Existing Automated Detection of Leakage in Fluorescein Angiography Images with Application to Malarial Retinopathy, a framework that can automatically detect three types of leakage (large focal, punctuate focal, and vessel segment leakage) and validated it on images from patients with malarial retinopathy whereas in the proposed work leakage in Diabetic Retinopathy is detected and the performance is validated.

In existing work only single level saliency map for malarial retinopathy FA images was detected whereas in proposed system a multiscale saliency is proposed for the detection of focal leakages in FA images. In table 1, the proposed method is based on two saliency cues: intensity and compactness features under multi-level super pixels. Then in table 2, gives the comparison of the saliency values of the super pixels at different levels are estimated in the intensity and compactness with SLIC and FCM SLIC methods respectively. While the intensity cue characterizes the intensity contrast among different super pixels, the compactness cue characterizes how densely (or sparsely) the salient pixels distribute inside a super pixel.

6. CONCLUSION

The automatic detection of the leakage presents various challenges. The leakages are hard to distinguish from background variations because it typically low contrast. Automatic detection of leakage in retinopathy images can be confused by other dark areas in the image such as the blood vessels. A framework was developed for automated detection of leakage and has tested it on FA images from patients with diabetic retinopathy. The framework benefits from two major components: saliency map generation, and leakage detection. The implemented method demonstrated satisfactory overall performancein FA images. These results depend in part on applying the concept of saliency to medical image processing. This framework has the potential to be developed further as a useful tool for fast, accurate and objective assessment of leak in a range of retinal diseases.

REFERENCES

  1. G. Richard, G. Soubrane, and L. Yanuzzi, Fluorescein and Icg Angiography: Textbook and Atlas, 2nd ed. New York:Thieme, 1998.
  2. T. Y. P. Chui et al., “Comparison of adaptive optics scanning light ophthalmoscopic fluorescein angiography and offset pinhole imaging,” Biomed. Opt. Exp., vol. 5, no. 4, pp. 1173–1189, 2014.
  3. N. Patton et al., “Retinal vascular image analysis as a potential screening tool for cerebrovascular disease: A rationale based on homology between cerebral and retinal microvasculatures,” J. Anatomy, vol. 206, no. 4, pp. 319–348, 2005.
  4. M. J. Potchen et al., “Acute brain MRI findings in 120 Malawian children with cerebral malaria: New insights into an ancient disease,” Am. J. Neuroradiol., vol. 33, no. 9, pp. 1740–1746, 2012.
  5. L. Itti, C. Koch, and E. Niebur, “A model of saliency-based visual attention for rapid scene analysis,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 20, no. 11, pp. 1254–1259, Nov. 1998.
  6. R. Achanta, S. Hemami, F. Estrada, and S. Susstrunk, “Frequencytuned salient region detection,” in Proc. IEEE CVPR, Jun. 2009, pp. 1597–1604.
  7. C. Koch and C. Ullman, “Shifts in selective visual attention: Towards the underlying neural circuitry,” Human Neurbiol., vol. 4, no. 4, pp. 219–227, 1985.
  8. M.-M. Cheng, G.-X. Zhang, N. J. Mitra, X. Huang, and S.-M. Hu, “Global contrast based salient region detection,” in Proc. IEEE CVPR, Jun. 2011, pp. 409–416.
  9. X. Chen, H. Zhao, P. Liu, B. Zhou, and W. Ren, “Automatic salient object detection via maximum entropy estimation,” Opt. Lett., vol. 38, no. 10, pp. 1727–1729, 2013.
  10. R. Achanta et al., “SLIC superpixels compared to state-of-the-art superpixel methods,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 34, no. 11, pp. 2274–2282, Nov. 2012.
  11. Y. Zhao et al., “Automated detection of leakage in fluorescein angiography images with application to malarial retinopathy,” Sci. Rep., vol. 5, p. 10425, Jun. 2015.
  12. H. Rabbani, M. J. Allingham, P. S. Mettu, S. W. Cousins, and S. Farsiu, “Fully automatic segmentation of fluorescein leakage in subjects with diabetic macular edema,” Invest. Ophthalmol. Vis. Sci., vol. 56, no. 3, pp. 1482–1492, 2015.
  13. L. Martínez-Costa et al., “Macular edema computer-aided evaluation in ocular vein occlusions,” Compute. Biomed. Res., vol. 31, no. 5, pp. 374–384, 1998.
  14. R. P. Phillips, T. Spencer, P. G. B. Ross, P. F. Sharp, and J. V. Forrester, “Quantification of diabetic maculopathy by digital imaging of the fundus,” Eye, vol. 5, no. 1, pp. 130–137, 1991.
  15. J. Saito, S. T. D. Roxburgh, D. Sutton, and A. Ellingford, “A new method of image analysis of fluorescein angiography applied to age-related macular degeneration,” Eye, vol. 9, no. 1, pp. 70–76, 1995.
  16. E. Trucco, C. R. Buchanan, T. Aslam, B. Dhillon, "Contextual detection of ischemic regions in ultra-wide-field-of-view retinal fluorescein angiograms",Proc. IEEE EMBC, pp. 6739-6742, Aug. 2007.
  17. C.-L. Tsai et al., "Automatic characterization of classic choroidal neovascularization by using AdaBoost for supervised learning",Invest. Ophthalmol. Vis. Sci., vol. 52, pp. 2767-2774, Apr. 2011.
  18. Y. Yuan, J. Wang, B. Li, M. Q.-H. Meng, "Saliency based ulcer detection for wireless capsule endoscopy diagnosis",IEEE Trans. Med. Imag., vol. 34, no. 10, pp. 2046-2057, Oct. 2015.
  19. D. Mahapatra, Y. Sun, "MRF-based intensity invariant elastic registration of cardiac perfusion images using saliency information",IEEE Trans. Biomed. Eng., vol. 58, no. 4, pp. 991-1000, Apr. 2011.

Ujjwal, K. S. Deepak, A. Chakravarty, J. Sivaswamy, "Visual saliency based bright lesion detection and discrimination in retinal images",Proc.IEEE ISBI, pp. 1436-1439, Apr. 2013.

1