About MEAJO | Editorial board | Search | Ahead of print | Current Issue | Archives | Instructions to authors | Online submission | Subscribe | Advertise | Contact | Login 
Middle East African Journal of Ophthalmology Middle East African Journal of Ophthalmology
Users Online: 5031   Home Print this page Email this page Small font sizeDefault font sizeIncrease font size


 
  Table of Contents 
ORIGINAL ARTICLE
Year : 2015  |  Volume : 22  |  Issue : 2  |  Page : 186-191  

Quantitative analysis of segmented fluorescein angiography images for the follow-up of choroidal neovascular membrane


1 Department of Ophthalmology, North Bengal Medical College, West Bengal, India
2 Department of Ophthalmology, RG Kar Medical College, Kolkata, West Bengal, India
3 Department of Electrical Engineering, IIT, New Delhi, India
4 Department of School of Medical Science and Technology, IIT, Kharagpore, West Bengal, India
5 Department of Ophthalmology, RIO, Kolkata, West Bengal, India

Date of Web Publication1-Apr-2015

Correspondence Address:
Sambuddha Ghosh
BB41/8, Salt Lake City, Kolkata - 700 064
India
Login to access the Email id

Source of Support: None, Conflict of Interest: None


DOI: 10.4103/0974-9233.151869

Rights and Permissions
   Abstract 

Purpose: The aim of this study was to evaluate choroidal neovascular (CNV) lesions with fluorescein angiography (FA) and to identify quantitative parameters and correlate these parameters to treatment outcomes.
Subjects and Methods: This institution based cross-sectional study evaluated 30 eyes with active sub-foveal predominantly classic CNV treated with bevacizumab. Pre- and post-injection segmented FA images were analyzed. Lesion area and CNV lesion were manually delineated. Outcome measure was the change 1-month after each injection in different intensity values (0-255 divided in eight regions A [lowest intensity] to H [highest intensity] on a linear scale) in lesion area, perimeter, greatest linear dimension (GLD), area, visual acuity (VA) and central macular thickness (CMT).
Results: At month 3, statistically significant changes from baseline occurred in VA, CMT, lesion area, GLD and perimeter (P < 0.05 all comparisons). Change in CMT from baseline to 3 months postinjection was correlated with change in VA (P = 0.009, r = 0.469) and intensity regions B (P = 0.001, r = −0.565), D (P = 0.001, r = 0.560), E (P = 0.035, r = 0.386). At month 3, change in intensity values 0-63 (A + B) was negatively correlated with CMT (P = 0.001, r = −0.575) and lesion area (P = 0.019, r = −0.427); change in intensity values 64-223 (C-G) was positively correlated with CMT (P = 0.000, r = 0.636) and lesion area (P = 0.002, r = 0.551).
Conclusions: Decrease in area, GLD, perimeter and area with intensity ≥ 64 on segmented FA were associated with a favorable outcome of treatment. These parameters may be useful adjuncts to existing evaluation techniques during follow-up of CNV.

Keywords: Age-related Macular Degeneration, Bevacizumab, Choroidal Neovascular Membrane, Fundus Fluorescein Angiography


How to cite this article:
Ghosh S, Haldar P, Ravindran P, Chatterjee J, Paranjape SV, Bhaduri G. Quantitative analysis of segmented fluorescein angiography images for the follow-up of choroidal neovascular membrane. Middle East Afr J Ophthalmol 2015;22:186-91

How to cite this URL:
Ghosh S, Haldar P, Ravindran P, Chatterjee J, Paranjape SV, Bhaduri G. Quantitative analysis of segmented fluorescein angiography images for the follow-up of choroidal neovascular membrane. Middle East Afr J Ophthalmol [serial online] 2015 [cited 2020 Aug 4];22:186-91. Available from: http://www.meajo.org/text.asp?2015/22/2/186/151869


   Introduction Top


Age-related macular degeneration (AMD) is the leading cause of irreversible vision loss in the elderly in developed western countries. [1] In the coming years, the prevalence of AMD in the Indian subcontinent is likely to follow a trend similar to the developed world. [2] Wet (exudative) AMD constitutes approximately 10% of AMD cases and accounts for 80% of the severe vision loss due to AMD. [3] The choice of treatment depends on the location, type and size of the lesion which is determined by fluorescein angiography (FA). An accurate angiographic distinction is of clinical importance for the choice of treatment as well as for follow-up of the lesion. [4] Intravitreal bevacizumab (off-label) injection has been proposed as an effective treatment for neovascular AMD. [5] Three consecutive monthly injections of anti-VEGF antibody followed by retreatment as-needed (PRN) is a commonly practiced treatment protocol. [6] However Schaal et al. reported a 50% decrease in bio-efficacy after approximately three intravitreal injections of bevacizumab in exudative AMD. [7] Currently, the treatment outcome is assessed primarily on the basis of visual acuity (VA) and optical coherence tomography (OCT) findings. However in cases with significant decrease in VA, or increase in fluid or hemorrhage on clinical examination, OCT alone is not always sufficient to detect lesion activity and repeat FA is advised to consider re-injection based on the evidence of leakage. [5] In a recent study, indocyanine green angiography has been proposed to help clinicians predict the change in choroidal neovascular (CNV) size and clinical course of neovascular AMD cases undergoing intravitreal bevacizumab therapy. However, majority of their cases were occult CNV cases (47.1%). [8]

Discrepancy between FA and high-resolution OCT in detection of macular disease has been reported in the literature. [9] In a study by Khurana et al., spectral-domain (SD-OCT) was found to be more effective than time domain (TD-OCT) to detect abnormalities in CNV with fluorescein leakage after anti-VEGF therapy. However, they reported failure to detect abnormalities in 41% and 10% of eyes with TD-OCT and SD-OCT respectively in the presence of detectable leakage on FA. [10] In such cases, during follow-up, subjective assessment of leakage on FA is sometimes difficult. Good inter-observer agreement (mean kappa coefficient = 0.64) regarding membrane type (a qualitative assessment) and high inter-observer variation (mean kappa coefficient = 0.40) regarding membrane size (a quantitative assessment) on FA in CNV has been reported. [11] Hyperfluorescence on FA may result both from leakages that are a sign of activity as well as from window defects. An objective assessment of leakage before deciding re-injection will be more preferable but necessitates identification of useful parameters for proper and accurate assessment. The region of CNV is characterized by sub-regions of varying intensities. Due to this nonuniform behavior, it becomes difficult to compare FA images at baseline and after treatment. Hence, image processing is essential for characterizing these nonuniform changes. Compared to nonsegmented image analysis, a digital image processing strategy through FA image feature segmentation provides a better opportunity. [12] The purpose of our study was to identify quantitative parameters for CNV lesions on segmented FA images and to correlate these with treatment outcome. In this study, pretreatment and posttreatment segmented FA images were analyzed.


   Subjects and methods Top


We retrospectively analyzed the records of 30 eyes of consecutive 30 patients who attended the retina clinic of a tertiary care center in eastern India. This study was approved by the Institutional ethics committee. This study adheres to the 1964 Declaration of Helsinki. All persons provided informed consent prior to inclusion in the study.

All eyes had active predominantly classic sub-foveal CNV due to AMD and were treated with three consecutive monthly injections of intravitreal bevacizumab followed by intravitreal injections as needed. All treatments were performed after the patient underwent a thorough informed consent procedure. All patients were treated between March 2009 and December 2010. The image processing was performed at the Department of the School of Medical Science and Technology at the Indian Institute of Technology, Kharagpore, India. Eyes with VA >6/18, previously treated with laser photocoagulation/intravitreal injection/photodynamic therapy and media opacity compromising image quality were excluded. Patients with glaucoma, diabetic retinopathy, any macular disorder other than AMD, scars in the macula, random blood sugar >140 mg/dl, any history of regular use of hypoglycemic agent were excluded. Patients with cerebrovascular or peripheral vascular events were not treated with intravitreal injection. In patients with bilateral active CNV, the eye with worse VA was included for analysis.

Patients were examined at baseline and at follow-up visits at the end of 1-3 months. Data collected at the baseline visit were: Best corrected distance VA measured with a Snellen's chart and converted to the log of the minimum angle of resolution (logMAR) equivalent, intraocular pressure, color fundus image, FA (TRC.50DX, Topcon Corp., Tokyo, Japan) with the angle of coverage 35° (performed by one certified ophthalmic photographer) and measurement of central macular thickness (CMT) by OCT (fast macular thickness scan, 3.2 mm macular thickness map, Stratus OCT, version 4.0.2; Carl Zeiss Meditec, Dublin, CA, USA) by a single observer. Bevacizumab (Avastin® ; Genentech, Inc., South San Francisco, CA) 1.25 mg/0.05 ml was injected through the pars plana using a 30-gauge needle at 0-2 months. At each follow-up visit, patients underwent a detailed ocular examination along with FA and OCT measurements.

The FA images were selected from the images of same time range (60-90 s). From these images, the final input image frame selection was based on matching the mean intensity value of retinal vessels by MATLAB (version 7.8.0.347 [R2009a], Mathworks, Natick, Massachusetts, USA). While matching the mean intensity of vessels, the variation in intensity of signal strength of surrounding tissue may influence variation in selected images. To overcome this we carried out segmentation of retinal vasculature (under supervision of an experienced ophthalmologist) and then computed the mean intensity.

Lesion area was defined in the study as the entire complex of lesion components. Lesion components were classic CNV and any of four angiographic features that could obscure the boundary of CNV. They include: (1) Blood that is visible on normal color photographs and thick enough to obscure normal choroidal fluorescence; (2) hypofluorescence due to pigment or fibrous tissue, or blood not visible on color fundus photographs; (3) serous detachment of retinal pigment epithelium; (4) scar from CNV that stains or blocks fluorescence. [13]

The lesion area was delineated using Paint software(Microsoft corporation, USA) by a retina specialist with 15 years of experience. The segmentation of this lesion area was performed by MATLAB software. A mask was then generated which was a binary image which is all 1's in the lesion area and 0 elsewhere. This mask was then multiplied with the original FA image to segment the region of the lesion [Figure 1]a-e]. The input image was an 8-bit gray level FA image. The intensities in these images are in the range 0-255 where 0 corresponds to dark pixel (no intensity) and 255 correspond to the white pixel (full intensity). [14] These intensities were divided on a linear scale into eight regions; (A) 0-31, (B) 32-63, (C) 64-95, (D) 96-127, (E) 128-159, (F) 160-191, (G) 192-223 and (H) 224-255. This division is such that the region A is the lowest intensity region and region H is the highest intensity region. The intensity areas within the delineated lesion area were then noted. Two masked observers independently assessed the leakage on the FA image. Agreement between their assessment of leakage with the presence of different intensity regions in that image was evaluated.
Figure 1: (a) Fluorescein angiography (FA) image at baseline (case 8). (b) Segmented FA image at baseline. (c) Segmented FA image at month 1. (d) Segmented FA image at month 2. (e) Segmented FA image at month 3

Click here to view


The segmented area was quantified by calculating the area, perimeter of the lesion region, greatest linear dimension (GLD) and texture content using MATLAB software in the following manner: Area is the number of pixels in the region, perimeter is the number of pixels in the circumference (or boundary) of the region, GLD is the maximum length, in pixels, of the line segment that can be drawn inside the region.

Primary outcome measure of this study was the change 1-month after each injection in the area, GLD, perimeter and different intensity areas of CNV lesion. Secondary outcome measures were change 1-month after each injection in VA and CMT.

Statistical analyses were performed using software SPSS, version 18.0 (IBM Corp., New York, NY, USA).


   Results Top


The study sample was comprised of 30 eyes of 30 patients with a mean age of 71.5 ± 6.2 years (range, 67-79 years). There were 21 (70%) male and 9 (30%) female. Mean baseline area of the CNV lesion was 150121.2 ± 173426.1 pixels. The area changed statistically significantly from baseline to 1-3 months to − 15875.7 pixels (P = 0.001), −33476.1 pixels (P = 0.000) and − 38947.3 pixels (P = 0.000) respectively. Mean baseline GLD was 480.7 ± 272.7 pixels. GLD changed statistically significantly from baseline to 1-3 months to − 41.7 pixels (P = 0.003), −77.2 pixels (P = 0.000) and − 93.6 pixels (P = 0.000) respectively. The mean baseline perimeter was 571.9 ± 649.9 pixels. Perimeter changed statistically significantly from baseline to 1-3 months to − 111.5 pixels (P = 0.024), −153.0 pixels (P = 0.002) and − 198.1 pixels (P = 0.001) respectively.

Of all intensities, at baseline, the highest mean area was for intensity region C and the lowest for intensity region G. There was no area within intensity region H in any of the images. At 3 months, the maximum decrease in mean area from baseline was noted for region C [Figure 2]. At the end of month 3, postinjection decrease from baseline in area of intensity was significant for region C, F and total area of regions C-H [P < 0.05 all cases; [Table 1]. Baseline mean logMAR VA was 1.24 ± 0.43. VA improved statistically significantly from baseline to 1-3 months to − 0.28, −0.33 and − 0.39 respectively (P = 0.000 each comparison). Mean baseline CMT was 365.7 ± 115 μ. CMT changed statistically significantly from baseline to 1-3 months to − 67.0μ, −96.9μ and − 100.3μ respectively (P = 0.000 each comparison).
Figure 2: Mean area of different intensities at each visit (X axis showing area in pixels) with power trend line

Click here to view
Table 1: Area of the different intensity components (in pixels) at baseline and at follow-up after treatment


Click here to view


Change in logMAR VA from baseline to 3 months was significantly correlated with change in CMT (P = 0.009, r = 0.469; Pearson's test). Change in GLD from baseline to 3 months was significantly correlated with change in area (P = 0.002, r = 0.554; Pearson's test), perimeter (P = 0.024, r = 0.417; Pearson's test), intensity region F (P = 0.000, r = 0.624; Pearson's test) and G (P = 0.003, r = 0.538) [Table 2]. Change in CMT from baseline to month 3 was correlated with change in logMAR VA (P = 0.009, r = 0.469; Pearson's test), intensity region B (P = 0.001, r = −0.565; Pearson's test), D (P = 0.001, r = 0.560; Pearson's test), E (P = 0.035, r = 0.386; Pearson's test). Increase in total area of region A and B (intensity 0-63) was significantly correlated with decrease in CMT (P = 0.001, r = −0.575; Pearson's test) and lesion area (P = 0.019, r = −0.427; Pearson's test) at month 3. Decrease in total area of region C-G (intensity 64-223) was significantly correlated with decrease in CMT (P = 0.000, r = 0.636; Pearson's test) and lesion area (P = 0.002, r = 0.551; Pearson's test) at month 3.
Table 2: Correlation of change from baseline at 3 months after treatment in areas of the different intensity components with change in best-corrected visual acuity and lesion parameters


Click here to view


Good inter-observer agreement on the assessment of leakage (mean kappa coefficient = 0.68) was noted between the two masked observers. The area outlined by two experts as leakage did correspond to image intensity ≥64 (regions C-H) in 97% and 95% cases respectively.


   Discussion Top


Significant improvement in VA after each injection, compared to baseline in our study concurs with the published literature. [5],[6],[15],[16] We found a significant decrease in CMT over the short term which also is in agreement with reports in the literature. [16],[17] We observed a reduction in area, perimeter and GLD. A previous study reported a reduction in area of CNV (with intravenous bevacizumab) on quantitative analysis of FA using EYENAV Software (RIRRC, Baltimore Maryland, USA). [18] In that study, however a 60 s frame was used for image analysis. [18] Selection of the frame based on degree of intravascular fluorescence (as in the current study) is considered better compared to frame selected on absolute time as the latter may be influenced by factors such as rapidity of injection and vascular anatomy. [12] Another study of intravitreal bevacizumab reported a significant reduction in GLD on FA at 3 months using incorporated measuring software (ImageNet 2000; Topcon, Tokyo, Japan). [16] None of these studies analyzed segmented images. Comparison of computerized analysis with traditional grading methods in the analysis of FA of CNV has been previously reported. [19] One study correlated the volume of various spaces on OCT with FA parameters in neovascular AMD and visual function. [20] This study attempted to correlate clinical outcomes of a specific treatment with changes in specified quantitative parameters on FA. We described units of FA parameters in pixels as is actually measured by the software, as the magnification factor in different camera differs. However, the magnification factor of the respective camera can be used to convert it to other units.

We split the pixel intensity values across the image into eight components. There was a smooth variation of intensities in FA images, which makes it difficult to categorize the image into different intensity regions. Our aim was to classify the input images into different regions based on the intensity values. Although the gray scale has 0-255 intensity levels, the human eye cannot identify so many different intensity levels. There is no available data on how many intensity levels a human eye can normally identify in the gray scale of FA. Our decision to divide the intensity into 8 components was arbitrary but based on our observation that the division into 4 intensity components resulted in three intensity components within the lesion area apart from the intensity corresponding to the blood vessels. After dividing the lesion area into 8 components, there are still 7 components within the lesion area apart from blood vessels. However, >8 created too many regions that made the visualization and calculation difficult. As the components of the lesion were defined in the study as classic CNV and four other angiographic features, we considered seven different intensity components within the lesion area as acceptable.

Currently, subjective assessment by a trained expert is the only method for assessing leakage on FA. We found good agreement between their subjective assessments with the presence of intensity region ≥64 on FA. In our study, we observed change in areas of different intensities. Areas of lower intensity (below 64) that increased in size after the 2 nd and 3 rd dose likely represent areas with window defects with no leakage of dye. A significant correlation between change in higher intensity regions (C-H) with favorable outcome measures such as decreased CMT indicates that this parameter could be used as an outcome measure in the future in addition to clinical and OCT based evaluation. However, the advent of OCT angiography may decrease the number of invasive FA procedures over time.

Diagnosis and follow-up of occult lesion always poses a challenge. We excluded occult CNV and included the classic variety only. We excluded occult CNV because our objective was to determine the quantifiable parameters for CNV, which are most evident in a classic variety only. Hence, occult lesions are not ideal for the study of these parameters. Our experience holds promise of further studies with occult lesion based on our results.

One weakness of this study was the semi-automated image analysis. Complete automated analysis is, of course, the final cherished goal. An appropriate algorithm for fully automated analysis can be developed by software experts only when all the data regarding quantitative optical parameters of all variety of tissues involved in the lesion area are determined. For this initial evaluation, we required clinical experts to identify the lesion area. The rest of the processing and analysis was done using MATLAB software. We do not advocate our method in this form for monthly clinical follow-ups. A clinician friendly fully automated analysis algorithm may be developed in the future for more useful application of this result. However in the present form these new parameters may be valuable for a clinical trial as an outcome measurement to evaluate the effect of any new treatment.

A limitation of this study was the use of TD-OCT instead of SD-OCT. However, our aim was to study FA parameters and not OCT parameters except CMT. Other limitations of this study include the short-term follow-up, small study population and the failure to use the ETDRS chart for VA. Despite the limitations, this study demonstrates that quantitative analysis of FA images by the segmentation technique gives useful information about active leakage and holds promise for future application as an adjunct to existing methods for follow-up of treatment in CNV.

 
   References Top

1.
Bressler NM, Bressler SB, Congdon NG, Ferris FL 3 rd , Friedman DS, Klein R, et al. Potential public health impact of Age-Related Eye Disease Study results: AREDS report no 11. Arch Ophthalmol 2003;121:1621-4.  Back to cited text no. 1
    
2.
Woo JH, Sanjay S, Au Eong KG. The epidemiology of age-related macular degeneration in the Indian subcontinent. Acta Ophthalmol 2009;87:262-9.  Back to cited text no. 2
    
3.
Ferris FL 3 rd , Fine SL, Hyman L. Age-related macular degeneration and blindness due to neovascular maculopathy. Arch Ophthalmol 1984;102:1640-2.  Back to cited text no. 3
    
4.
Bressler NM, Bressler SB, Fine SL. Age-related macular degeneration. Surv Ophthalmol 1988;32:375-413.  Back to cited text no. 4
    
5.
Subramanian ML, Ness S, Abedi G, Ahmed E, Daly M, Feinberg E, et al. Bevacizumab vs ranibizumab for age-related macular degeneration: Early results of a prospective double-masked, randomized clinical trial. Am J Ophthalmol 2009;148:875-82.e1.  Back to cited text no. 5
    
6.
CATT Research Group, Martin DF, Maguire MG, Ying GS, Grunwald JE, Fine SL, et al. Ranibizumab and bevacizumab for neovascular age-related macular degeneration. N Engl J Med 2011;364:1897-908.  Back to cited text no. 6
[PUBMED]    
7.
Schaal S, Kaplan HJ, Tezel TH. Is there tachyphylaxis to intravitreal anti-vascular endothelial growth factor pharmacotherapy in age-related macular degeneration? Ophthalmology 2008;115:2199-205.  Back to cited text no. 7
    
8.
Rush RB, Rush SW, Aragon AV 2 nd , Ysasaga JE. Evaluation of choroidal neovascularization with indocyanine green angiography in neovascular age-related macular degeneration subjects undergoing intravitreal bevacizumab therapy. Am J Ophthalmol 2014;158:337-44.  Back to cited text no. 8
    
9.
Kozak I, Morrison VL, Clark TM, Bartsch DU, Lee BR, Falkenstein I, et al. Discrepancy between fluorescein angiography and optical coherence tomography in detection of macular disease. Retina 2008;28:538-44.  Back to cited text no. 9
    
10.
Khurana RN, Dupas B, Bressler NM. Agreement of time-domain and spectral-domain optical coherence tomography with fluorescein leakage from choroidal neovascularization. Ophthalmology 2010;117:1376-80.  Back to cited text no. 10
    
11.
Friedman SM, Margo CE. Choroidal neovascular membranes: Reproducibility of angiographic interpretation. Am J Ophthalmol 2000;130:839-41.  Back to cited text no. 11
    
12.
Walsh AC, Updike PG, Sadda SR. Quantitative fluorescein angiography. In: Schachat AP, Ryan SJ, editors. Retina. 4 th ed., Vol. 2. Philadelphia: Elsevier Mosby; 2006.  Back to cited text no. 12
    
13.
Bressler NM, Bressler SB, Fine SL. Neovascular (Exudative) age related macular degeneration. In: Schachat AP, Ryan SJ, editors. Retina. 4 th ed., Vol. 2. Philadelphia: Elsevier Mosby; 2006.  Back to cited text no. 13
    
14.
González RC, Woods RE. Digital Image Processing. 3 rd ed. New Jersey: Prentice Hall; 2008.  Back to cited text no. 14
    
15.
Emerson MV, Lauer AK, Flaxel CJ, Wilson DJ, Francis PJ, Stout JT, et al. Intravitreal bevacizumab (Avastin) treatment of neovascular age-related macular degeneration. Retina 2007;27:439-44.  Back to cited text no. 15
    
16.
Bashshur ZF, Haddad ZA, Schakal A, Jaafar RF, Saab M, Noureddin BN. Intravitreal bevacizumab for treatment of neovascular age-related macular degeneration: A one-year prospective study. Am J Ophthalmol 2008;145:249-56.  Back to cited text no. 16
    
17.
Melamud A, Stinnett S, Fekrat S. Treatment of neovascular age-related macular degeneration with intravitreal bevacizumab: Efficacy of three consecutive monthly injections. Am J Ophthalmol 2008;146:91-5.  Back to cited text no. 17
    
18.
Nguyen QD, Shah SM, Hafiz G, Do DV, Haller JA, Pili R, et al. Intravenous bevacizumab causes regression of choroidal neovascularization secondary to diseases other than age-related macular degeneration. Am J Ophthalmol 2008;145:257-66.  Back to cited text no. 18
    
19.
Chakravarthy U, Walsh AC, Muldrew A, Updike PG, Barbour T, Sadda SR. Quantitative fluorescein angiographic analysis of choroidal neovascular membranes: Validation and correlation with visual function. Invest Ophthalmol Vis Sci 2007;48:349-54.  Back to cited text no. 19
    
20.
Sadda SR, Liakopoulos S, Keane PA, Ongchin SC, Msutta S, Chang KT, et al. Relationship between angiographic and optical coherence tomographic (OCT) parameters for quantifying choroidal neovascular lesions. Graefes Arch Clin Exp Ophthalmol 2010;248:175-84.  Back to cited text no. 20
    


    Figures

  [Figure 1], [Figure 2]
 
 
    Tables

  [Table 1], [Table 2]



 

Top
  
 
  Search
 
    Similar in PUBMED
   Search Pubmed for
   Search in Google Scholar for
 Related articles
    Access Statistics
    Email Alert *
    Add to My List *
* Registration required (free)  

 
  In this article
    Abstract
   Introduction
   Subjects and methods
   Results
   Discussion
    References
    Article Figures
    Article Tables

 Article Access Statistics
    Viewed1958    
    Printed57    
    Emailed0    
    PDF Downloaded98    
    Comments [Add]    

Recommend this journal