CC BY
1
Development of a Method of Feature Space Formation for Assessment of Choroidea Condition from Retinal Angio-OCT Images
Nataly Yu. Ilyasova1'2*, Ravil T. Samigullin2, and Nikita S. Demin1,2
1 IPSI, NRC "Kurchatov Institute", 151 Molodogvardeyskaya, Samara 443001, Russian Federation
2 Samara National Research University, 34 Moskovskoye Shosse, Samara 443086, Russian Federation *e-mail: [email protected]
Abstract. This paper presents a technique for selecting regions of interest in retinal angio-OCT images to quantitatively analyze choroidea parameters in order to detect ocular diseases. The significance of this research lies in the fact that choroideas are an important part of the eye responsible for nourishing the retina and maintaining its normal function. Disruptions in the choriodea lead to various eye conditions, including degenerative retina diseases and glaucoma. A method for assessing choriodeal conditions has been developed based on identifying areas without vascular signal in retinal angiographic OCT images. Additional indicators for estimating choriodean status have been proposed. Comparative analysis of parameters between norm and pathological classes was conducted. The results of this work may be useful for ophthalmologists and contribute to improved diagnosis and treatment for eye conditions. © 2024 Journal of Biomedical Photonics & Engineering.
Keywords: biomedical images; optical coherence tomography images; threshold processing; choroid; quantitative features; quantitative analysis.
Paper #9118 received 21 Jun 2024; revised manuscript received 19 Sep 2024; accepted for publication 20 Sep 2024; published online 29 Sep 2024. doi: 10.18287/JBPE24.10.030306.
1 Introduction
The choroidea is a capillary layer that extends along Bruch's membrane and lies adjacent to the inner vasculature. It plays an important role in supporting the vessels of the retinal pigment epithelium and the outer part of the retina. The choroideas are the main vascular layers of the eye and influence the development of various ocular diseases, such as age-related macular degeneration, polypoidal chorioretinal vasculopathy, central serous choroidopathy, and myopic macular degeneration. The choroids also provide nutrition and waste removal from the retinas, playing an essential role in their health and function. Histological studies have confirmed that pathological processes affect the structure and vascularization of the choroid [1].
Diseases in which changes in choroidal thickness are observed include central serous chorioretinopathy, Vogt-Koyanagi-Harada disease, polypoidal choroidal vasculopathy, and age-related exudative macular degeneration.
Central serous chorioretinopathy (CSC) is characterised by the accumulation of fluid under the retina in the central region of the macula, leading to distortion of the visual field and decreased visual acuity. Changes in the thickness of chorioidea are associated with impaired blood supply and drainage in this area [2].
Vogt-Koynagi-Harad disease (VKH) is a rare inflammatory condition that affects the eyes, skin and other organs. The main symptoms of VKH include inflammation of the eye, changes in skin pigmentation and hair, and visual disturbances. Changes in the choroidal thickness of the retina are associated with an inflammatory process and immune mechanisms that affect blood vessels in the eye [3].
Polypoidal choroidal vasculopathy (PCV) is a type of age-related macular degeneration (AMD) in which abnormal vascular formations resembling polyps develop in the choroid. These polyps can lead to bleeding, swelling, and visual impairment. Changes in choroid thickness in PCV are associated with the formation and regression of these abnormal polyps [4].
Exudative AMD is a form of AMD in which abnormal vascular neoplasms accumulate under the retina, causing a decrease in visual function and distortion of the central field of vision. Changes in the thickness of the choroid in AMD are associated with growth of these tumors and disruption of normal vascular structure [5].
The study of these diseases and choroidal changes helps to expand our understanding of the pathophysiology of ocular pathologies and develop new approaches to diagnosis and treatment. The concept of choroidal vascular index (VI) was introduced to assess the degree of vascular density in the choroid and can serve as an additional tool for monitoring these conditions [6].
Investigations of ocular diseases require accurate and reliable diagnostic methods [7-9]. In recent years, optical coherence tomography (OCT) has become a key tool for evaluating the choroid and other structures of the eye [10-12]. OCT is a non-invasive high-resolution imaging technique that produces cross-sectional tissue slices with micrometre resolution [13]. Unlike other methods, such as fluorescence angiography or indocyanine green angiography, which have gained importance in the diagnosis of ocular diseases, OCT provides more detailed and objective information about the structure and thickness of the choroid [14].
The main features of OCT include the ability to image layers of eye tissues and assess their thickness with high resolution [15, 16]. OCT allows clear visualization of the choroid, detection of changes and determination of various pathological conditions such as choroidal neoplasms. It also monitors the effectiveness of treatment and evaluates the prognosis of disease. Due to its ability to obtain three-dimensional images, OCT provides a more comprehensive study of the structure and changes of the choroids compared to other methods.
OCT images of the ocular fundus were used to develop algorithms for glaucoma diagnosis registration in Refs. [10, 11]. Also, OCT images of the ocular fundus from [12] were studied and used to determine retinal edema. A classification system was developed based on these studies. Similar studies have been conducted in Refs. [15-21].
The study method chosen was the analysis and processing of ocular fundus angiography images.
2 Materials and Methods
The task of ocular fundus angiography-OCT image processing is to form sets of features for quantitative analysis of the choroidea. To develop a technology for assessing the state of the choroid by identifying areas where vascular signal is absent. The following types of choroidal OCT angiography (angiography) images are used as input data for processing. Angio-OCT images have a size of 3*3 mm2.
OCT scans with angiography (OCTA), as shown in Fig. 1(a, b), are used to assess the choriocapillary status. The choriocapsular status is determined by the presence of voids in the scans that are larger than 5000 ^m2
(200 pixels, 1 pixel equals 25 ^m2). The reviewed literature concludes that people with central serous chorioretinopathy have a greater number of large voids (equal to or greater than 5000 ^m2) compared to healthy volunteers [22]. Also Fig. 1(a) shows an image without pathology and Fig. 1(b) shows an image with pathology. It can be seen that the images differ in their brightness. Therefore, we proposed a criterion for estimating the transparency of the choroid layer based on image brightness.
(a)
(b)
Fig. 1 Angio-OCT of the choroidea: (a) normal, (b) pathology.
For the study, images were taken from 9 to 18 ^m deep into the vasculature, from Bruch's membrane, and strata of 9 ^m thickness were experimentally selected. Images at these depths were selected based on the fact that the number of hades is least in images at those parameters [22].
The method itself consists of subjecting OCTA scans to local thresholding using the Phansalkar algorithm, where voids are defined as pixels of black color [23]. A sign to track pathological changes is an increase in the number of large voids or an increase in their size [22].
To further analyze the choroidal index, we selected relevant features on which a detailed analysis was performed. The technology for evaluating diagnostic features to detect pathologies from retina angio-OCT images is presented below.
3 Methods for Evaluating Diagnostic Features for Quantitative Choroideal Analysis
To evaluate the images, textual features need to be extracted [24].
We selected the following features to quantify the choroidea on retinal angio-OCT images:
1) the number of voids (K1),
2) the area of voids that are equal to or greater than 5000 ^m2 (£2),
3) the contour characteristic (K3),
4) the transparency coefficient (K4).
Transparency is characterized by a positive offset of
the mean value of brightness with respect to the median value:
Ic=-
(1)
'max
I
2
For opaque media, Ic. The quantitative expression of this parameter is the transparency coefficient. The coefficient is determined by Eq. (2) [25] :
KA =
(J- Ic)
(2)
where K4 is the transparency coefficient, I is the mean luminance value, Ic is the median luminance value.
K3 is a contour feature, which is defined as the sum of all contour pixels in the image. First, we select the contours using openCV and then we use Eq. (3):
K,
= 2LiZr=i/(¿J),
(3)
where K3 is the number of pixels representing contours; h is the height of the image, w is the width of the image, the function defining the pixel color I(i, j) = {1 if pixel (i, j) is white, 0 otherwise}.
For choroidea optical coherence tomography (OCT) angiography images, criteria have been proposed such as the number and area of voids. The presence of voids equal to or greater than 5000 ^m2 determines the choriocapillary status. Individuals with choriocappillary pathology have a higher number of larger voids compared to healthy patients, according to exploratory studies.
A technology has been developed for processing angio-OCT images.
4 Technology for Choroidea Assessment
To assess the choroidea, a technique was developed consisting of the following steps including pretreatment methods:
a) median filter processing,
b) threshold processing,
c) extreme filter processing,
d) allocation of voids,
e) an assessment of their quantitative characteristics.
The first step is to apply the median filter with a window size of 3*3. The result of image processing by the median filter is shown in Fig. 2 (b).
Thresholding was applied locally using the Phansalkar algorithm, where voids are defined as black-colored pixels [23]. The unique threshold was calculated using the Eq.:
treshold (x,y) = M (l + pe-qM + l)), (4)
where (x,y) are the pixel coordinates, M is the mean of all pixels within the radius, SD is the standard deviation of all pixels within the window, p is a constant that the exponential term affects the threshold - for low values of
p (0 < p < 1) it behaves like Sauvola's method, too high (p > 5) and too many background pixels can be classified as foreground pixels (default p = 3), q - constant, and the algorithm behaves like Sauvola's algorithm above a certain values (default k = 0.25), r - dynamic range of standard deviation (default r = 0.5). The result of the thresholding process is shown in Fig. 2 (c). To sharpen and improve image quality, an extreme filter was used, described by the following Eq.:
g
= \fi'
fi, npu fo-fl< fN- fo fN, npu fo- fi > fN-fo,
(5)
where fi are the numbers of the variation series of the population with N numbers, g is the output value.
(a)
(b)
(c)
(d) (e)
Fig. 2 Voids extraction: (a) original image, (b) median filter result, (c) local thresholding result, (d) extreme filter result, (e) voids image.
Fig. 3 Scheme of the algorithm of voids selection [16].
Table 1 Static characteristics for "normal" and "pathology" classes. Feature group Norma Patalogy
Parameter
K1
K2 K3 K4
„,„ , . Mathematical „-„c . . Mathematical RMS deviation , ,. RMS deviation , ,. _expectation_expectation
588.174 6.786 885.687 0.074
7302.2 6,4 818.7 0.08
217.459 4.962 2613.857 0.035
7935.8 67.2 16105.1 -0.619
(b)
Fig. 4 Visualisation of the foramen: (a) normal, (b) pathology.
The result of applying the extreme filter is shown in Fig. 2 (d).
The selection of voids is based on the criterion of four-connectivity, i.e., only samples adjacent to the sides of an object are considered neighbors. The algorithm consists in processing the image matrix line by line. An arbitrary sample («1, «2) is selected, if it
belongs to the background and is equal to zero, then the next sample is processed. If it does not belong to the background (i.e., it is equal to one), then its belonging to any object is analyzed. For this purpose, two neighboring samples already processed are taken into account. Depending on the values of these neighboring samples, the current sample may be attached to an object. If neighboring samples belong to background, a new object is created [25]. The algorithm is shown in Fig. 3.
Experimental investigations of the proposed technique on natural images are presented below.
5 Analysis of the Results of Image Processing of Chorioid Angio-OCT Images
For the analysis, we took images at depths of 9-18 ^m from the Bruch's membrane, with a step size of 9 ^m.
Figs. 5-8 show the distribution graphs of the obtained values of the signs.
(a)
1.6E-03
1.4E-03
1.2E-03
ë
lyi C 1.0E-03
id
x> 8.0E-04
o
3 6.0E-04
.D
4.0E-04
i/i
b 2.0E-04
0.0E+00
- - - Pathology
1
l'
1
1
/\ '
/ 1
/ 1 i
2000 4000 6000 Number of voids
8000 10000
4.5E-04
4.0E-04
c 3.5E-04
01
■o c 3.0E-04
o
2.5E-04
-O
IA 2.0E-04
a
1.5E-04
1.0E-04
5.0E-05
0.0E+00
1 _ __ Pathology
Norma
l
H
-1- /
\— N
_J. Li -- i - - -1
Fig. 5 Void distribution density plot.
-5000 0 5000 10000 15000 20000 25000 30000 Contour characterisation
0.09
>- 0.08
c at 0.07
xt
c 0.06
o
a 0.05
-Q
s 0.04
Q 0.03
0.02
0.01
0.00
---Patalogy
-Norma I\
1 1 i i
1 * 1 1
1 \
l 1 M
l 1 1 I
1 V
! i V
Fig. 7 Contour characteristic distribution density plot.
-20 0 20 40 60 80
Number of voids with area => 5000 (im2
100
Fig. 6 Density plot of the distribution of voids with an area equal to or greater than 5000 ^m2.
10
9 8 7 6 5 4 3 2 1 0
---Pathology
t \ Norma
r* _¡M ._ 1
1 1
1
Ï 1 1
—1— 1 1_
1 1
1 1 —1— \
—1— J
-0.6 -0.4 -0.2 0
Transparency factor
0.2
0.4
Fig. 8 Transparency ratio distribution graph.
15000 10000 5000 0
• Norma . Pat h l-.l fiflU
A t
I M
L-
20 40 60
Number Df voids with area -i 5000 (im?
(a)
0.3
0.2
0.1
s 0
>- -0.1
c OJ -0.2
fi si -0.3
-0.4
H -0.5
-0.6
-0.7
* s,
• Norm a
A Pathology *4
*
• •
Number of voids
8000 100ÛÛ
(b)
• Norma
• • A Pathol ogy
* a
I -0.2
I -0.3
25000
° 20000
t; 15000
"5 10000
S
I 5000 u
• Norma A Pathology
A
T
2000 4000 6000 8000 10Û00 Number of voids
(c)
• Norma ▲ Pathology *
••
_A_
S
Number of voids with area >= 5000 urn2 (d)
2000 4000 6000 8000 10000 Number of voids
(e)
Fig. 9 Graphs of distribution of objects of two classes in the space of Contour characteristic : (a) for the number of voids with area greater than 5000 ^m2, (b) the ratio of the number of voids with area greater than 5000 ^m2 to the number of all voids, (c) for the number of all voids; Transparency coefficient: (d) for the number of voids greater than 5000 ^m2, (e) for the number of all voids.
This choice was based on previous studies, which showed that the lowest number of voids were found in formations at these depths [22]. Fig. 4 shows a visualisation of voids for normal and pathological conditions.
According to the above graphs, we can already say that there is a clear separation between the classes of normal and pathological objects. We also plotted the distribution of objects in the two classes, and the results are presented in Fig. 9. Table 1 shows the static characteristics of features for the "normal" and "pathology" classes.
6 Conclusion
This paper presents a technique for selecting regions of interest in retinal angio-OCT images for quantitative analysis of choroide parameters to detect various ocular diseases. In this work, features of the choroid were
References
studied and algorithms for extracting parameters and counting were reviewed.
During the study, a method was developed to assess the condition of the choroids by finding areas with no vascular signal (voids) in angio- OCT images of the choroidal region, as well as a set of features was formed for quantitative assessment of choroid: contour characteristics, transparency coefficient, number of voids, and number of void areas greater than or equal to 5000 дт2. The results obtained were analyzed. On the angio-OCT images, the contour and transparency coefficients showed clear separation between classes.
Acknowledgment
The work was carried out within the state assignment of IPSI, NRC "Kurchatov institute".
Disclosures
The authors declare that they have no conflict of interest.
1. R. Agrawal, P. Gupta, K. A. Tan, C. M. G. Cheung, T. Y. Wong, and C. Y. Cheng, "Choroidal vascularity index as a measure of vascular status of the choroid: Measurements in healthy eyes from a population-based study," Scientific Reports 6, 21090 (2016).
2. L. I. Balashevich, M. V. Gatsu, E. M. Kasimov, and N. G. Iskanderova, "Current views on the etiology and pathogenesis of central serous chorioretinopathy," Ophthalmosurgery 2, 63-67 (2007). [in Russian]
3. A. Nurmukhamedova, A. R. Utaralina, N. A. Khvaleva, and A. S. Dobrynin, "Vogt-Koyanagi-Harada disease," Proceedings of Student Research, G. Yu. Gulyaev (Ed.), Penza, Russian Federation, 201-203 (2021). ISBN: 978-500159-742-1. [in Russian]
4. E. A. Abdulaeva, "Polypoidal choroidal vasculopathy," Kazan Medical Journal 98(3), 403-409 (2017). [in Russian]
5. S. E. Avetisov, V. P. Yerichev, M. V. Budzinskaya, and M. A. Karpilova, "Age-related macular degeneration and ocular hypertension," National Journal Glaucoma (1), 62-67 (2013). [in Russian]
6. K. Breher, L. Terry, T. Bower, and S. Wahl, "Choroidal Biomarkers: A Repeatability and Topographical Comparison of Choroidal Thickness and Choroidal Vascularity Index in Healthy Eyes," Translational Vision Science & Technology 9(11), 8 (2020).
7. N. Yu. Ilyasova, N. S. Demin, A. S. Shirokanev, A. V. Kupriyanov, and E. A. Zamytsky, "Method for selection macular edema region using optical coherence tomography data," Computer Optics 44(2), 250-258 (2020).
8. N. Yu. Ilyasova, D. V. Kirsh, and N. S. Demin, "Decision-making support system for the personalisation of retinal laser treatment in diabetic retinopathy," Computer Optics 46(5), 774-782 (2022).
9. R. A. Paringer, A. V. Mukhin, N. Yu. Ilyasova, and N. S. Demin, "Neural networks application for semantic segmentation of fundus," Computer Optics 46(4), 596-602 (2022). [in Russian]
10. H. Raja, M. U. Akram, S. G. Khawaja, M. Arslan, A. Ramzan, and N. Nazir, "Data on OCT and fundus images for the detection of glaucoma," Data in Brief 29, 105342 (2020).
11. G. An, K. Omodaka, K. Hashimoto, S. Tsuda, Y. Shiga, N. Takada, T. Kikawa, H. Yokota, M. Akiba, and T. Nakazawa, "Glaucoma diagnosis with machine learning based on optical coherence tomography and colour fundus images," Journal of Healthcare Engineering 2019, 4061313 (2019).
12. B. Hassan, R. Ahmed, B. Li, O. Hassan, and T. Hassan, "Automated retinal edema detection from fundus and optical coherence tomography scans," in 2019 5th International Conference on Control, Automation and Robotics (ICCAR), IEEE, 325-330 (2019).
13. A. Uji, S. Balasubramanian, J. Lei, E. Baghdasaryan, M. Al-Sheikh, and S. R. Sadda, "Choriocapillaris imaging using multiple en face optical coherence tomography angiography image averaging," JAMA Ophthalmology 135(11), 1197-1204 (2017).
14. Z. Chu, H. Zhou, Y. Cheng, Q. Zhang, and R. K. Wang, "Improving visualisation and quantitative assessment of choriocapillaris with swept source OCTA through registration and averaging applicable to clinical systems," Scientific Reports 8(1), 16826 (2018).
15. R. F. Spaide, "Choriocapillaris flow features follow a power law distribution: implications for characterisation and mechanisms of disease progression," American Journal of Ophthalmology 170, 58-67 (2016).
16. K. A. Halavataya, K. V. Kozadaev, and V. S. Sadau, "Adjusting videoendoscopic 3D reconstruction results using tomographic data," Computer Optics 46(2), 246-251 (2022).
17. M. A. Burnasheva, A. N. Kulikov, and D. S. Maltsev, "Artifact-free evaluation of choriocapillaris perfusion in central serous chorioretinopathy," Vision 5(1), 3 (2020).
18. Z. Chu, Q. Zhang, G. Gregori, P. J. Rosenfeld, and R. K. Wang, "Guidelines for imaging the choriocapillaris using OCT angiography," American Journal of Ophthalmology 222, 92-101 (2021).
19. K. Takayama, H. Kaneko, Y. Ito, K. Kataoka, T. Iwase, T. Yasuma, T. Matsuura, T. Tsunekawa, H. Shimizu, A. Suzumura, E. Ra, T. Akahori, and H. Terasaki, "Novel classification of early-stage systemic hypertensive changes in human retina based on OCTA measurement of choriocapillaris," Scientific Reports 8(1), 15163 (2018).
20. F. F. Conti, V. L. Qin, E. B. Rodrigues, S. Sharma, A. V. Rachitskaya, J. P. Ehlers, and R. P. Singh, "Choriocapillaris and retinal vascular plexus density of diabetic eyes using split-spectrum amplitude decorrelation spectral-domain optical coherence tomography angiography," British Journal of Ophthalmology 103(4), 452-456 (2019).
21. O. Loria, L. Kodjikian, P. Denis, C. Vartin, S. Dimassi, L. Gervolino, A. Maignan, R. Kermarrec, C. Chambard, P. Pradat, and T. Mathis, "Quantitative analysis of choriocapillaris alterations in swept-source optical coherence tomography angiography in diabetic patients," Retina 41(9), 1809-1818 (2021).
22. D. S. Maltsev, A. V. Fomin, A. N. Kulikov, and A. S. Vasiliev, "Evaluation of the choriocapillaris status using highspeed spectral optical coherence tomography with angiography function and image averaging technology," Vestnik Ophthalmologii 137(3), 76-84 (2021). [in Russian]
23. Z. Chu, Y. Cheng, Q. Zhang, H. Zhou, Y. Dai, Y. Shi, G. Gregori, P. J. Rosenfeld, and R. K. Wang, "Quantification of Choriocapillaris with Phansalkar Local Thresholding: Pitfalls to Avoid," American Journal of Ophthalmology, 213, 161-176 (2020).
24. A. A. Kolchev, D. V. Pasynkov, I. A. Egoshin, I. V. Kliouchkin, and O. O. Pasynkova, "Classification of benign and malignant solid breast lesions on the ultrasound images based on the textural features: the importance of the perifocal lesion area," Computer Optics 48(1), 157-165 (2024).
25. N. Y. Ilyasova, A. V. Kupriyanov, and A. G. Khramov, Information technologies of image analysis in the tasks of medical diagnostics, A. S. Bugaev (Ed.), Radio and Communication, Moscow (2012). ISBN: 5-89776-014-4. [in Russian]
