November 2015
Volume 56, Issue 12
Free
Cornea  |   November 2015
Quantifying Ocular Surface Inflammation and Correlating It With Inflammatory Cell Infiltration In Vivo: A Novel Method
Author Affiliations & Notes
  • Giulio Ferrari
    Cornea and Ocular Surface Disease Unit Eye Repair Lab, IRCCS San Raffaele Scientific Institute, Milan, Italy
  • Alessandro Rabiolo
    Cornea and Ocular Surface Disease Unit Eye Repair Lab, IRCCS San Raffaele Scientific Institute, Milan, Italy
  • Fabio Bignami
    Cornea and Ocular Surface Disease Unit Eye Repair Lab, IRCCS San Raffaele Scientific Institute, Milan, Italy
  • Federico Sizzano
    Flow Cytometry Resource Advanced Cytometry Technical Applications Laboratory, IRCCS San Raffaele Scientific Institute, Milan, Italy
    Flow Cytometry Core Facility, Nestlè Institute of Health Sciences, Lausanne, Switzerland
  • Alessio Palini
    Flow Cytometry Resource Advanced Cytometry Technical Applications Laboratory, IRCCS San Raffaele Scientific Institute, Milan, Italy
    Flow Cytometry Core Facility, Nestlè Institute of Health Sciences, Lausanne, Switzerland
  • Chiara Villa
    Flow Cytometry Core Facility, Nestlè Institute of Health Sciences, Lausanne, Switzerland
  • Paolo Rama
    Cornea and Ocular Surface Disease Unit Eye Repair Lab, IRCCS San Raffaele Scientific Institute, Milan, Italy
  • Correspondence: Giulio Ferrari, Cornea and Ocular Surface Disease Unit, Eye Repair Lab, IRCCS San Raffaele Scientific Institute, Via Olgettina 60, 20132 Milan, Italy; ferrari.giulio@hsr.it
  • Footnotes
     Current affiliation: *Nestlè Institute of Health Sciences, Lausanne, Switzerland.
Investigative Ophthalmology & Visual Science November 2015, Vol.56, 7067-7075. doi:10.1167/iovs.15-17105
  • Views
  • PDF
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to Subscribers Only
      Sign In or Create an Account ×
    • Get Citation

      Giulio Ferrari, Alessandro Rabiolo, Fabio Bignami, Federico Sizzano, Alessio Palini, Chiara Villa, Paolo Rama; Quantifying Ocular Surface Inflammation and Correlating It With Inflammatory Cell Infiltration In Vivo: A Novel Method. Invest. Ophthalmol. Vis. Sci. 2015;56(12):7067-7075. doi: 10.1167/iovs.15-17105.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose: The purpose of this study was to develop a novel, objective, and semiautomated method to quantify conjunctival redness by correlating measured redness with standard clinical redness and symptom scales and inflammatory cell infiltration.

Methods: Eleven outpatients presenting with mild to severe conjunctival hyperemia were included in the study. Clinical examination included patient history; visual analogue score (VAS) for ocular symptoms; 25-item National Eye Institute Visual Function Questionnaire (NEI-VFQ 25) for quality of life/vision; photographs of the anterior segment graded for conjunctival hyperemia, using Efron, relative redness of image (RRI), and edge feature (EF) scales; and conjunctival impression cytology analyzed by flow cytometry. Differences between affected and unaffected eyes were evaluated, and correlations among questionnaire scores, ocular hyperemia grading scores, and assessment of biological markers were performed.

Results: Visual analogue score (P < 0.0001), Efron scale (P = 0.0003), RRI scores (P = 0.0004), and EF scores (P < 0.0001) and the percentage of granulocytes (defined as cluster of differentiation [CD] 45dim; P = 0.0080) were significantly higher in affected eyes. Conversely, the percentage of CD45bright leukocytes was reduced in affected eyes (P = 0.0054). Both the RRIs and EFs were positively correlated with VAS, Efron scale, percentages of conjunctival granulocytes, and CD45brightCD3neg cells, whereas they were negatively correlated with the percentage of CD45brightCD3pos cells. Edge feature and RRI were correlated (Spearman r = 0.78, P < 0.0001).

Conclusions: Ocular redness is a cardinal sign driving clinical judgment in highly prevalent ocular disorders; hence, we suggest that our semiautomated and reproducible method may represent a helpful tool in the follow-up of these patients.

Italian Abstract

Among the four cardinal signs heralding inflammation (i.e., rubor, tumor, dolor, and calor), rubor (i.e., redness) of the bulbar conjunctiva is the most relevant driver of clinical judgment in cases of ocular surface disorders. Objective and repeatable grading is key in the follow-up of a number of highly prevalent and disabling disorders, including keratitis,1 uveitis,2 dry eye,3 and others.4 This is particularly relevant in clinical trials, where standardized and consistent endpoint measurements are highly desirable. A number of methods have been proposed to measure conjunctival redness. Among these, we can differentiate between manual qualitative methods, or grading scales, and semi- or fully automatic methods. 
In grading scales, a score is arbitrarily given depending on the number, density, and tortuosity of vessels. The Efron,5 validated bulbar redness,6 and McMonnies scales are among the scales most frequently used.7 Despite their intuitiveness, clinical grading scales exhibit extreme variability among different investigators8 as well as for the same observer over time.9 
In order to overcome the limits of qualitative measurements, several (semi-) automated techniques have been described. Most of them are based on a combination of color quantification,1013 edge detection,11 and fractal analysis.14 Although they quantify conjunctival hyperemia objectively and reproducibly, such methods have not widely spread into clinical practice due to the requirement of dedicated instruments and/or trained operators, setup costs, accessibility, and amount of time needed to analyze images. 
To overcome these pitfalls, we developed a method to objectively quantify ocular hyperemia by using instruments commonly found in ophthalmic outpatient clinics (i.e., a slit-lamp unit and a computer) in a simple and low-cost way. The algorithm we propose here quantifies ocular redness by detection of edge feature (EF) and relative color extraction (i.e., relative redness of the image [RRI]). Both of the algorithms assign a number from 0 to 1, to each pixel of the slit-lamp image. A value of zero is assigned to a pixel having “no red,” whereas a value of 1 is assigned to a “red” pixel. This method does not require human intervention in the grading, hence, variability is limited to image capture, which can be easily reduced by setting standardized parameters (slit-lamp beam, light intensity, and others). For this reason, it could be easily used in the follow-up of patients, thus reducing interoperator variability and assessment of patient status by different physicians over time. 
In order to validate our method, we correlated results with those obtained with manual grading scales (Efron) or questionnaires to assess ocular symptoms (visual analogue score [VAS])15 and the impact of the disease on quality of vision/life (25-Item National Eye Institute Visual Function Questionnaire [NEI-VFQ 25]).16,17 In addition, we assessed whether our results correlated with objective signs of inflammation, specifically leukocyte infiltration, measured by flow cytometry performed using conjunctival impression cytology samples. 
Methods
Study Population
A total of eleven patients affected by mild to severe conjunctival hyperemia were included in this prospective observational study. Patient characteristics are reported in the Table. Mean (±SD [standard deviation]) patient age was 60.0 ± 14.5 (range, 35–79) years, and there were six males and five females. The healthy eye of 5 of the 11 subjects served as an internal control (no conjunctival hyperemia). Mean age (±SD) of internal controls was 60.8 ± 12.7 (range, 44–76) years; two were male and three were female. No significant age differences between patients and controls was appreciated (P = 0.9172). The study was conducted at the Cornea and Ocular Surface Disease Unit, San Raffaele Hospital, Milan, Italy, in compliance with the Declaration of Helsinki and approved by the local ethical committee. Informed consent was obtained. Inclusion criteria were conjunctival hyperemia ≥ 2 according to Efron scale for conjunctival redness5; and patient age ≥ 18 years old. Exclusion criteria were patient was clinically judged at risk for corneal perforation and patient was unable to give informed consent. Enrolled patients were evaluated at baseline (day 0 [D0]) and 20 ± 8 (range, 7–35) days later (D1). The examination included patient history, questionnaires to assess ocular symptoms and quality of vision/life, photographs of the anterior segment, and conjunctival impression cytology. Both eyes were examined at all time points, even if only one eye fit the inclusion criteria. Two patients (nos. 10 and 11) were lost at follow-up. 
Table
 
Demographic Features
Table
 
Demographic Features
Questionnaires
Two copies of the VAS questionnaire were administered to patients, one for each eye, in order to assess ocular symptoms including foreign body sensation, burning/stinging, itching, pain, stick feeling, blurred vision, and photophobia.15 
An Italian, validated version of the 25-Item National Eye Institute Visual Function Questionnaire (NEI-VFQ 25) was administered to patients in order to assess the impact of the disease on quality of vision/life.16,17 
Photograph Acquisition
Two images were acquired for each eye by the same operator (AR). Patients were asked to look temporally and nasally in order to assess nasal and temporal bulbar conjunctiva. Eyelids were held open to reveal the entire cornea and the maximum amount of bulbar conjunctiva.10 Slit-lamp parameters were set as follows: white light with application of the diffuser, magnification ×10, maximum slit width, angle of slit-lamp arm of 45°, and maximum light intensity at one-half. Room lights were switched on. Photographs were stored through the Phoenix version 2.1 software (OPW, Hodgkin, IL, USA) as JPEG (Joint Photographic Experts Group) images with a resolution of 1624 × 1232 pixels. Conjunctival hyperemia was graded by two methods, a manual semiquantitative method and a semiautomatic quantitative method. 
Manual Grading
Manual grading of the anonymized images was performed by an expert clinician (GF) using the Efron scale for conjunctival redness.5 The Efron scale consists of five images having progressive degrees of ocular hyperemia. A printed color version of the scale is displayed for evaluation by the clinician with no time limit for each image. All images were manually graded within a single session to control for changes in room illumination and/or monitor brightness and contrast. 
Semiautomatic Grading
All images were processed using ImageJ software (http://imagej.nih.gov/ij/; provided in the public domain by the National Institutes of Health, Bethesda, MD, USA). Conjunctival hyperemia was quantified by a second clinician (AR) using two digital indices, namely RRI12 and EF. Both the RRI and EF indices are described by adimensional numbers ranging from 0 to 1. (See Supplementary Fig. S1 for key passages to digitally quantify conjunctival hyperemia.) The maximum amount of bulbar conjunctiva was included in the analyses, excluding everything but conjunctiva itself, such as cornea, eyelids, or eyelashes. For that purpose, a region of interest (ROI) was drawn using freehand selection around the exposed conjunctiva, and all but the ROI was replaced by a pure white background by using the “clear outside” function. Two different algorithms were applied to the resulting image in order to calculate RRI and EF. 
Software Algorithms
Relative redness of image was calculated as described by Papas12 and divided by the total number of pixels, specifically:  where i and j are pixel coordinates; Rij, Gij, and Bij are red, green, and blue intensities, respectively, for a pixel at positions i and j in the image array; and NoP is the total number of image pixels. In order to calculate RRI index, we wrote an ImageJ macro to (1) exclude pure white background (RGB code 255.255.255); (2) exclude pixels with specular reflection, defined by R, G, and B values above 220 (Ref. 10); (3) to extract R, G, and B values for each pixel; and (4) to calculate RRI.  
As described by Fieguth and Simpson,11 EF was calculated as the ratio between the number of edge pixels, computed by Canny edge detection algorithm,18 and the total number of conjunctival pixels.    
To calculate EF: (1) ImageJ function “Find edges” was launched; (2) pictures were split into the three color channels; (3) the green channel was selected because it provided the best signal-to-noise ratio (SNR); (4) “canny edge detector” plugin (provided in the public domain by http://rsbweb.nih.gov/ij/plugins/canny/index.html) was launched, and plugin function “Conn Thresholding” was selected; (5) a threshold value was manually defined for every picture; (6) plugin function “Hysteresis” was used in order to get binary black and white pictures, where edge pixels were the white ones, whereas black pixels were nonedge plus background ones; (7) in order to exclude background from the computation, images were manually cut and pasted into a pure blue background (RGB code 000.000.255); and (8) an ImageJ macro was written to exclude background pixels and to calculate the number of edge, nonedge, and the edge-to-total pixels ratio. 
Sample Collection for Flow Cytometry
To avoid any interference in flow cytometric analysis, sample collection was performed prior to fluorescein application. Superficial conjunctival cells were collected through impression cytology as previously described.19,20 Briefly, 1 drop oxybuprocaine hydrochloride 0.4% was instilled in patients' eyes; after 10 seconds, sterile nitrocellulose membranes (Merck Millipore Ltd., Etobicoke, ON, Canada), previously divided into two specular semicircles, were gently applied on both side of the filter to each eye onto the unexposed bulbar conjunctiva, superotemporally, inferotemporally, superonasally, and inferonasally for approximately 20 seconds. Filters were then immediately placed in 15 mL sterile tubes containing 5 mL complete medium (RPMI medium plus 10% fetal bovine serum) and kept at 4°C. Then, samples were moved into a thermic bag, carried to the Flow Cytometry Core Facility and processed within 24 hours. 
Flow Cytometry Analysis
Briefly, tubes were vortexed for 30 seconds, and filter disks were removed using a sterile forceps. Then, cells were pelleted by centrifugation for 5 minutes at 300g. Supernatant was discarded, and 50 μL dilution mixture (PBS with 1% bovine serum albumin) containing Syto-16 (final concentration 2.5 μM; Life Technologies, Monza, Italy) and monoclonal antibodies were added to the pellet for 20 minutes at room temperature. Antibodies included CD45-phycoerhythrin-cyanine 7 (PE-Cy7), HLA-DR-allophycocyanin (APC), and CD14-electron coupled dye (ECD) (all Beckman Coulter, Milan, Italy), whereas CD16-APC-H7 and CD3-Pacific Blue were from Becton-Dickinson (Milan, Italy). CD15 Alexa-Fluor 700 (Campoverde, Milan, Italy) was kindly donated by G Oliveira (San Raffaele Scientific Institute, Milan, Italy). All antibodies were properly titrated using blood samples in order to reduce background in the multicolor environment. At the end of incubation, 150 μL PBS and propidium iodide (PI; 0.5 μg/mL final concentration; Sigma-Aldrich, Milan, Italy) were added. Samples were acquired using a LSR Fortessa cell analyzer equipped with 355-, 405-, 488-, 561-, and 640-nm laser lines (Becton Dickinson). To check instrument performance, in order to ensure robustness and reproducibility of the data, calibrator beads (8-peaks rainbow beads; Spherotech, Lake Forest, IL, USA) were used at the beginning of each experimental session. Gating strategy was based on the exclusion of dead cells (positive for PI and Syto-16, PIpos/Syto16pos) and debris (positive or negative for PI and negative for Syto16, PIpos-neg/Syto16neg) from the analysis. After gating for doublet exclusion (“Singlet” gate), we characterized cells as CD45dim or CD45bright on the basis of fluorescence intensity of CD45 staining as observed during flow cytometry. CD45dim defines the whole granulocyte population as being CD14dim/neg and CD16pos and CD15pos as well. CD45bright included lymphocytes (CD3posHLA-DRpos/neg) and monocytes/macrophages, defined as CD3neg, CD14bright, or CD16pos (see Supplementary Fig. S2 for the visualization of gating strategy). All data were stored in a list mode file (Flow Cytometry Standard version 3.1) and were analyzed using FCS Express 4 software (DeNovo Software, Glendale, CA, USA). Data were expressed as the percentage of the population of interest in the parental gate (i.e., %CD45dim in the “Singlet” gate). 
Statistical Analysis
All data were expressed as means ± standard error of the mean (SEM). All measurements (first and second visits) were pooled. Differences between affected and unaffected eyes were assessed using t-test and Mann-Whitney U test for parametric and nonparametric variables, respectively. Relationship between variables was investigated using Spearman correlation tests. A P value less than 0.05 was considered significant. Prism 5.0 software (GraphPad Software, San Diego, CA, USA) was used to analyze the data. 
Results
Clinical Data
Patients' eyes at D0 examination are shown in Figure 1. Ocular signs and conjunctival hyperemia grading scores in affected and unaffected eyes are shown in Figure 2. Visual analogue score questionnaire scores (Fig. 2A) were 33.47 ± 3.60 and 3.54 ± 3.22 in affected and unaffected eyes, respectively (P < 0.0001). NEI-VFQ 25 score was 40.37 ± 4.58 (data not shown). Conjunctival hyperemia assessed with both clinical and digital score systems was significantly greater in affected eyes than in unaffected ones. Specifically, Efron scores for conjunctival hyperemia (Fig. 2B) were 2.79 ± 0.23 and 0.90 ± 0.23 in affected and unaffected eyes, respectively (P = 0.0003). With regard to digital scores, RRIs (Fig. 2C) were 0.46 ± 0.01 and 0.40 ± 0.01 in affected and unaffected eyes, respectively (P = 0.0004); and EFs (Fig. 2D) were 0.25 ± 0.02 and 0.08 ± 0.01, respectively (P < 0.0001). As shown in Figure 3, Efron score strongly correlated with both RRI (Spearman r = 0.93, P < 0.0001) and EF (Spearman r = 0.81, P < 0.0001). Edge feature and RRI correlated with each other (Spearman r = 0.78, P < 0.0001) (Fig. 3C). NEI-VFQ25 negatively correlated with patients' worst VAS scores (Spearman r = −0.52, P = 0.0195) (data not shown). No further correlations were appreciated between NEI-VFQ 25 and VAS, RRI, or EF (data not shown). Finally, no significant differences were found between D0 and D1 in any of the parameters considered. 
Figure 1
 
Baseline clinical views (magnification ×10) of eyes included in the study. Unaffected eyes were #1 left, #3 right, #7 left, #8 left, and #11 right.
Figure 1
 
Baseline clinical views (magnification ×10) of eyes included in the study. Unaffected eyes were #1 left, #3 right, #7 left, #8 left, and #11 right.
Figure 2
 
Statistical comparison of VAS (A), Efron scale for conjunctival redness (B), RRI (C), and EF (D) scores between affected (n = 31) and unaffected eyes (n = 9). Mann-Whitney U test was used to assess differences between the two groups for factors (BD), whereas unpaired t-test was used for (A). ***Significant at P < 0.001; ****Significant at P < 0.0001; error bars indicate SEM.
Figure 2
 
Statistical comparison of VAS (A), Efron scale for conjunctival redness (B), RRI (C), and EF (D) scores between affected (n = 31) and unaffected eyes (n = 9). Mann-Whitney U test was used to assess differences between the two groups for factors (BD), whereas unpaired t-test was used for (A). ***Significant at P < 0.001; ****Significant at P < 0.0001; error bars indicate SEM.
Figure 3
 
The digital parameters relative redness index and edge features were positively correlated with the Efron scale (A, B) and to each other (C). Number of samples = 40; ****significant at P < 0.0001; r = Spearman's rank correlation coefficient.
Figure 3
 
The digital parameters relative redness index and edge features were positively correlated with the Efron scale (A, B) and to each other (C). Number of samples = 40; ****significant at P < 0.0001; r = Spearman's rank correlation coefficient.
Flow Cytometry Results
Baseline impression cytology data of one affected eye (patient 9, right eye) were excluded from analysis due to insufficient sample retrieval. Supplementary Figure S2 shows gating strategies used to determine cellular populations. CD45bright and CD45dim leukocyte percentages were significantly different between the two groups. Specifically, CD45dim cells were 51.10% ± 6.84% and 18.62% ± 10.09% (P = 0.0080), whereas CD45bright leucocytes were 37.05% ± 6.03% and 72.45% ± 9.78% (P = 0.0054) in affected and unaffected eyes, respectively. No significant differences between the two groups were appreciated in CD45brightCD3pos T lymphocytes (P = 0.3084), CD45brightCD3neg cells (P = 0.2492), activated CD45brightCD3posHLADRpos T lymphocytes (P = 0.4481), CD45brightCD3negHLA-DRpos cells (P = 0.6980), or CD45brightCD3negCD14bright, CD16pos monocytes (P = 0.3525). Flow cytometry results, expressed as percentages, in affected and unaffected eyes are shown in Figure 4
Figure 4
 
Impression cytology cell populations are different between inflamed (n = 30) and noninflamed (n = 9) eyes. Statistical comparison of CD45dim and CD45bright leukocytes (A), CD45brightCD3pos and CD45brightCD3neg cells (B), activated T and non–T cells defined as HLA-DRpos cells (C), and monocytes (D). Mann-Whitney U test was used to assess differences between the two groups. **Significant at P < 0.01; error bars indicate SEM.
Figure 4
 
Impression cytology cell populations are different between inflamed (n = 30) and noninflamed (n = 9) eyes. Statistical comparison of CD45dim and CD45bright leukocytes (A), CD45brightCD3pos and CD45brightCD3neg cells (B), activated T and non–T cells defined as HLA-DRpos cells (C), and monocytes (D). Mann-Whitney U test was used to assess differences between the two groups. **Significant at P < 0.01; error bars indicate SEM.
Correlation Analysis
VAS and digital hyperemia grading systems were significantly correlated (Fig. 5). Specifically, a positive correlation was appreciated between ocular symptoms and conjunctival hyperemia, namely between VAS score and both RRI (Spearman r = 0.53, P = 0.0004) and EF (Spearman r = 0.62, P < 0.0001) (Figs. 5A, 5B). Moreover, ocular symptoms correlated directly with the presence of CD45dim granulocytes (Spearman r = 0.46, P = 0.0031) (Fig. 5C) and inversely with CD45bright cells (Spearman r = −0.35, P = 0.0298) (Fig. 5D). None of the other flow cytometry markers were correlated with VAS score. Figures 6 and 7 show correlation between conjunctival hyperemia, expressed as RRI and EF scores, and flow cytometry markers. The granulocyte infiltrate (Figs. 6A, 6B) correlated directly with both RRI (Spearman r = 0.69, P < 0.0001) and EF (Spearman r = 0.64, P < 0.0001), whereas total conjunctival CD45bright leukocytes were inversely related with RRI (Spearman r = −0.57, P = 0.0002) (Fig. 6C) and EF (Spearman r = −0.58, P < 0.0001) (Fig. 6D). Among CD45bright leukocytes, T lymphocytes were inversely correlated with RRI (Spearman r = −0.41, P = 0.0105) and EF (Spearman r = −0.43, P = 0.0064) (Figs. 7A, 7B), while CD45brightCD3neg cells directly correlated with RRI (Spearman r = 0.39, P = 0.0147) (Fig. 7C) and EF (Spearman r = 0.42, P = 0.0075) (Fig. 7D). The remaining cell populations were not significantly correlated to RRI or EF. No correlation was appreciated between NEI-VFQ 25 and any cell population (data not shown). 
Figure 5
 
Relative redness index and edge features and CD45dim and CD45bright cells are directly correlated with visual analogue scale. Correlation between (A) VAS and RRI, (B) VAS and EF, (C) VAS and granulocytes, and (D) VAS and CD45bright cells. Number of samples = 39; *significant at P < 0.05; **significant at P < 0.01; ***significant at P < 0.001; ****significant at P < 0.0001; r = Spearman's rank correlation coefficient.
Figure 5
 
Relative redness index and edge features and CD45dim and CD45bright cells are directly correlated with visual analogue scale. Correlation between (A) VAS and RRI, (B) VAS and EF, (C) VAS and granulocytes, and (D) VAS and CD45bright cells. Number of samples = 39; *significant at P < 0.05; **significant at P < 0.01; ***significant at P < 0.001; ****significant at P < 0.0001; r = Spearman's rank correlation coefficient.
Figure 6
 
CD45dim and CD45bright cells are correlated with RRI and EF. Correlation between (A) RRI and granulocytes, (B) EF and granulocytes, (C) RRI and CD45brigh cells, and (D) EF and CD45brigh cells. Number of samples = 39; ***significant at P < 0.001; **** significant at P < 0.0001; r = Spearman's rank correlation coefficient.
Figure 6
 
CD45dim and CD45bright cells are correlated with RRI and EF. Correlation between (A) RRI and granulocytes, (B) EF and granulocytes, (C) RRI and CD45brigh cells, and (D) EF and CD45brigh cells. Number of samples = 39; ***significant at P < 0.001; **** significant at P < 0.0001; r = Spearman's rank correlation coefficient.
Figure 7
 
CD45brightCD3pos and CD45brightCDneg correlated to RRI and EF. Correlation between (A) RRI and T cells, (B) EF and T cells, (C) RRI and CD45brightCD3neg cells, and (D) EF and CD45brightCD3neg cells. Number of samples = 39; *significant at P < 0.05; **significant at P < 0.01; r = Spearman's rank correlation coefficient.
Figure 7
 
CD45brightCD3pos and CD45brightCDneg correlated to RRI and EF. Correlation between (A) RRI and T cells, (B) EF and T cells, (C) RRI and CD45brightCD3neg cells, and (D) EF and CD45brightCD3neg cells. Number of samples = 39; *significant at P < 0.05; **significant at P < 0.01; r = Spearman's rank correlation coefficient.
Discussion
In this paper, we report a semiautomatic method to objectively quantify ocular hyperemia using instruments commonly found in every ophthalmic outpatient clinic (i.e., a slit-lamp unit and a computer) in a simple and low-cost way. Among all proposed strategies to quantify conjunctival redness, edge detection (i.e., EF) and relative color extraction (i.e., RRI) appeared to be the most stable, reliable, and sensitive; moreover, such techniques were highly correlated with clinical grading scales.21 Peterson et al.21 proposed that edge detection and relative color extraction resemble the clinical perception of conjunctival hyperemia, which mostly relies on vessels coverage area and ocular redness, in a more objective and reliable way. 
Furthermore, to increase the correlation with robust and well-described biological markers of ocular surface inflammation, we evaluated the amount and phenotype of conjunctival inflammatory cell infiltration by means of impression cytology and flow cytometry. Our results confirm a solid increase in granulocytes in inflamed eyes, as reported by Williams et al.,22 although CD45bright cells were significantly reduced. The percentages of monocytes and T lymphocytes were not significantly different between the two groups, although monocytes were most represented in inflamed eyes and T lymphocytes in noninflamed eyes. Because flow cytometry data are expressed as percentages, CD45bright cells were probably reduced in inflamed eyes due to the concomitant increase of CD45dim granulocytes. Additionally, it is well known that granulocytes are more represented than lymphocytes and monocytes in acute corneal inflammation, which affected the patients recruited in this study, and that they play a crucial role in pathogen elimination and resolution of inflammation.23 
In order to test whether our findings might have relevance in a clinical setting, such as a clinical trial, we searched for correlations between clinical redness indices and infiltrating inflammatory cells. Interestingly, ocular redness was positively correlated with granulocytes and CD45brightCD3neg non–T cells, whereas total CD45bright cells and CD45brightCD3pos T lymphocytes exhibited a negative correlation. The CD45brightCD3neg population contains monocytes that, despite the absence of significant correlation, are increased in affected eyes and could play a role in active inflammation, together with other CD3neg cells such as NK cells. Indeed, the presence of NK cells was instrumental to the development of maximal ocular surface inflammation.24 On the other hand, we found that the percentage of T lymphocytes was inversely related to ocular redness indices. This could be due to the presence of regulatory T cells, which are known for their immunomodulatory activity. In fact, this specific lymphocyte subset has been associated with reduction of inflammation.25 Such correlations between ocular redness and cytofluorimetric markers was statistically significant also when the Efron scale was used. We further analyzed our data, creating homogenous diagnostic groups, specifically, ocular cicatricial pemphigoid and bacterial ulcer. Cicatricial pemphigoid patients (two patients with both eyes affected) showed higher percentages of CD45dim cells than CD45bright cells (cumulative results for the median of the two eyes at two time points: CD45dim: 85.12% versus those in CD45bright: 6.17%). Interestingly, the majority of CD45bright cells were CD3neg (cumulative results for the median of the two eyes at the two time points: CD3neg: 95% vs. CD3pos: 5%). Thus, a small population of CD3pos cells is present in the conjunctiva of pemphigoid patients. Indeed, T lymphocytes have a significant role in the disease, as described before.26 In the two patients suffering from corneal bacterial ulcers we noticed a prevalence of CD45dim at the first time point in the affected eye (patient 9: 95.59%; patient 11: 88.09%), whereas at the second time point the percentage of CD45dim decreased for patient 9 to 31.53% (patient 11 was lost to follow-up), whereas CD3pos cells increased to 66.67%. Although the limited sample does not allow a definitive conclusion, this could reflect a switch in the immune response toward a Th1/Th2 phenotype (extensively studied in mouse models by Hazlett and Hendricks27) after an infiltration of polymorphonucleated cells in a very early phase of the infection. Indeed, the presence of lymphocytes has been described in viral, but also bacterial keratitis.28 The Th1/Th2 switch could explain the weaker correlation between CD3 and redness indexes. In fact, these cells could actively stimulate inflammation or, on the contrary, play a role in immune modulation. Further studies are needed to address additional phenotypic characterization, to better dissect the role of T lymphocyte subpopulations and other non–T-cell subsets (monocytes, NK cells) in the setting of conjunctival hyperemia. 
A potential limitation of our grading, which is shared by all other computerized methods, is the time required to process images, which is greater than that needed to complete clinical grading scales. We estimated that in this study, the time needed to take the eye pictures and to analyze them with both RRI and EF was approximately 5 minutes. We think that time can be reduced when the operator gains experience, and is short enough to keep the procedure feasible. Another potential limitation is represented by some source of residual variability, in particular regarding image acquisition at the slit-lamp; conjunctival ROI selection; and choice of threshold values for EF (not for RRI). Many studies, including ours, have tried to minimize variability in image acquisition by setting fixed slit-lamp parameters.10,11,13 In addition, Amparo et al.10 proposed an elegant white-balance correction algorithm to balance differences in lightening conditions. With regard to conjunctival ROIs selection, small interoperator differences should be unimportant given the large amount of total conjunctiva and the fact that the indices do not depend on the number of pixels. 
In summary, we report a novel, semiautomated method to quantify ocular surface inflammation. This quantification requires a slit-lamp and a personal computer, which are generally available in any clinical setting. Two indices are generated as an output, and they are significantly correlated with the percentages of CD45dimgranulocytes, total CD45bright cells, CD45brightCD3pos T lymphocytes and CD45brightCD3neg non–T cells. Because conjunctival redness is a key sign of inflammation and significantly drives clinical judgment in highly prevalent ocular disorders, we suggest that this method may represent a useful tool in the follow-up of these disorders. Additionally, it could be used as a robust endpoint measure in clinical trials testing anti-inflammatory treatments of the ocular surface. 
Acknowledgments
Disclosure: G. Ferrari, None; A. Rabiolo, None; F. Bignami, None; F. Sizzano, None; A. Palini, None; C. Villa, None; P. Rama, None 
References
Rowe AM, St Leger AJ, Jeon S, Dhaliwal DK, Knickelbein JE, Hendricks RL. Herpes keratitis. Prog Retin Eye Res. 2013; 32: 88–101.
Chang JH, Wakefield D. Uveitis: a global perspective. Ocul Immunol Inflamm. 2002; 10: 263–279.
Hessen M, Akpek EK. Dry eye: an inflammatory ocular disease. J Ophthalmic Vis Res. 2014; 9: 240–250.
Leibowitz HM. The red eye. N Engl J Med. 2000; 343: 345–351.
Efron N. Grading scales for contact lens complications. Ophthalmic Physiol Opt. 1998; 18: 182–186.
Schulze MM, Jones DA, Simpson TL. The development of validated bulbar redness grading scales. Optom Vis Sci. 2007; 84: 976–983.
McMonnies CW, Chapman-Davies A. Assessment of conjunctival hyperemia in contact lens wearers. Part II. Am J Optom Physiol Opt. 1987; 64: 251–255.
Terry RSD, Wong R, Papas E. Variability of clinical investigators in contact lens research. Optom Vis Sci. 1995; 72: 16.
Chong TST, Fonn D. The repeatability of discrete and continuous anterior segment grading scales. Optom Vis Sci. 2000; 77: 244–251.
Amparo F, Wang H, Emami-Naeini P, Karimian P, Dana R. The ocular redness index: a novel automated method for measuring ocular injection. Invest Ophthalmol Vis Sci. 2013; 54: 4821–4826.
Fieguth P, Simpson T. Automated measurement of bulbar redness. Invest Ophthalmol Vis Sci. 2002; 43: 340–347.
Papas EB. Key factors in the subjective and objective assessment of conjunctival erythema. Invest Ophthalmol Vis Sci. 2000; 41: 687–691.
Rodriguez JD, Johnston PR, Ousler GW,III Smith LM, Abelson MB. Automated grading system for evaluation of ocular redness associated with dry eye. Clin Ophthalmol. 2013; 7: 1197–1204.
Schulze MM, Hutchings N, Simpson TL. The use of fractal analysis and photometry to estimate the accuracy of bulbar redness grading scales. Invest Ophthalmol Vis Sci. 2008; 49: 1398–1406.
Ohashi Y, Ebihara N, Fujishima H, et al. A randomized, placebo-controlled clinical trial of tacrolimus ophthalmic suspension 0.1% in severe allergic conjunctivitis. J Ocul Pharmacol Ther. 2010; 26: 165–174.
Mangione CM, Lee PP, Gutierrez PR, et al. Development of the 25-item National Eye Institute Visual Function Questionnaire. Arch Ophthalmol. 2001; 119: 1050–1058.
Rossi GC, Milano G, Tinelli C. The Italian version of the 25-item National Eye Institute Visual Function Questionnaire: translation validity, and reliability. J Glaucoma. 2003; 12: 213–220.
Canny J. A computational approach to edge detection. IEEE Trans Comput. 1986; 6: 679–698.
Brignole-Baudouin F, Ott AC, Warnet JM, Baudouin C. Flow cytometry in conjunctival impression cytology: a new tool for exploring ocular surface pathologies. Exp Eye Res. 2004; 78: 473–481.
Epstein SP, Gadaria-Rathod N, Wei Y, Maguire MG, Asbell PA. HLA-DR expression as a biomarker of inflammation for multicenter clinical trials of ocular surface disease. Exp Eye Res. 2013; 111: 95–104.
Peterson RC, Wolffsohn JS. Objective grading of the anterior eye. Opt Vis Sci. 2009; 86: 273–278.
Williams GP, Tomlins PJ, Denniston AK, et al. Elevation of conjunctival epithelial CD45INTCD11b(+)CD16(+)CD14(−) neutrophils in ocular Stevens-Johnson syndrome and toxic epidermal necrolysis. Invest Ophthalmol Vis Sci. 2013; 54: 4578–4585.
Kolaczkowska E, Kubes P. Neutrophil recruitment and function in health and inflammation. Nat Rev Immunol. 2013; 13: 159–175.
Reyes NJ, Mayhew E, Chen PW, Niederkorn JY. NKT cells are necessary for maximal expression of allergic conjunctivitis. Int Immunol. 2010; 22: 627–636.
Sehrawat S, Suvas S, Sarangi PP, Suryawanshi A, Rouse BT. In vitro-generated antigen-specific CD4+ CD25+ Foxp3+ regulatory T cells control the severity of herpes simplex virus-induced ocular immunoinflammatory lesions. J Virol. 2008; 82: 6838–6851.
Elder MJ, Lightman S. The immunological features and pathophysiology of ocular cicatricial pemphigoid. Eye (Lond). 1994; 8 (pt 2) : 196–199.
Hazlett LD, Hendricks RL. Reviews for immune privilege in the year 2010: immune privilege and infection. Ocul Immunol Inflamm. 2010; 18: 237–243.
Kwon B, Hazlett LD. Association of CD4+ T cell-dependent keratitis with genetic susceptibility to Pseudomonas aeruginosa ocular infection. J Immunol. 1997; 159: 6283–6290.
Figure 1
 
Baseline clinical views (magnification ×10) of eyes included in the study. Unaffected eyes were #1 left, #3 right, #7 left, #8 left, and #11 right.
Figure 1
 
Baseline clinical views (magnification ×10) of eyes included in the study. Unaffected eyes were #1 left, #3 right, #7 left, #8 left, and #11 right.
Figure 2
 
Statistical comparison of VAS (A), Efron scale for conjunctival redness (B), RRI (C), and EF (D) scores between affected (n = 31) and unaffected eyes (n = 9). Mann-Whitney U test was used to assess differences between the two groups for factors (BD), whereas unpaired t-test was used for (A). ***Significant at P < 0.001; ****Significant at P < 0.0001; error bars indicate SEM.
Figure 2
 
Statistical comparison of VAS (A), Efron scale for conjunctival redness (B), RRI (C), and EF (D) scores between affected (n = 31) and unaffected eyes (n = 9). Mann-Whitney U test was used to assess differences between the two groups for factors (BD), whereas unpaired t-test was used for (A). ***Significant at P < 0.001; ****Significant at P < 0.0001; error bars indicate SEM.
Figure 3
 
The digital parameters relative redness index and edge features were positively correlated with the Efron scale (A, B) and to each other (C). Number of samples = 40; ****significant at P < 0.0001; r = Spearman's rank correlation coefficient.
Figure 3
 
The digital parameters relative redness index and edge features were positively correlated with the Efron scale (A, B) and to each other (C). Number of samples = 40; ****significant at P < 0.0001; r = Spearman's rank correlation coefficient.
Figure 4
 
Impression cytology cell populations are different between inflamed (n = 30) and noninflamed (n = 9) eyes. Statistical comparison of CD45dim and CD45bright leukocytes (A), CD45brightCD3pos and CD45brightCD3neg cells (B), activated T and non–T cells defined as HLA-DRpos cells (C), and monocytes (D). Mann-Whitney U test was used to assess differences between the two groups. **Significant at P < 0.01; error bars indicate SEM.
Figure 4
 
Impression cytology cell populations are different between inflamed (n = 30) and noninflamed (n = 9) eyes. Statistical comparison of CD45dim and CD45bright leukocytes (A), CD45brightCD3pos and CD45brightCD3neg cells (B), activated T and non–T cells defined as HLA-DRpos cells (C), and monocytes (D). Mann-Whitney U test was used to assess differences between the two groups. **Significant at P < 0.01; error bars indicate SEM.
Figure 5
 
Relative redness index and edge features and CD45dim and CD45bright cells are directly correlated with visual analogue scale. Correlation between (A) VAS and RRI, (B) VAS and EF, (C) VAS and granulocytes, and (D) VAS and CD45bright cells. Number of samples = 39; *significant at P < 0.05; **significant at P < 0.01; ***significant at P < 0.001; ****significant at P < 0.0001; r = Spearman's rank correlation coefficient.
Figure 5
 
Relative redness index and edge features and CD45dim and CD45bright cells are directly correlated with visual analogue scale. Correlation between (A) VAS and RRI, (B) VAS and EF, (C) VAS and granulocytes, and (D) VAS and CD45bright cells. Number of samples = 39; *significant at P < 0.05; **significant at P < 0.01; ***significant at P < 0.001; ****significant at P < 0.0001; r = Spearman's rank correlation coefficient.
Figure 6
 
CD45dim and CD45bright cells are correlated with RRI and EF. Correlation between (A) RRI and granulocytes, (B) EF and granulocytes, (C) RRI and CD45brigh cells, and (D) EF and CD45brigh cells. Number of samples = 39; ***significant at P < 0.001; **** significant at P < 0.0001; r = Spearman's rank correlation coefficient.
Figure 6
 
CD45dim and CD45bright cells are correlated with RRI and EF. Correlation between (A) RRI and granulocytes, (B) EF and granulocytes, (C) RRI and CD45brigh cells, and (D) EF and CD45brigh cells. Number of samples = 39; ***significant at P < 0.001; **** significant at P < 0.0001; r = Spearman's rank correlation coefficient.
Figure 7
 
CD45brightCD3pos and CD45brightCDneg correlated to RRI and EF. Correlation between (A) RRI and T cells, (B) EF and T cells, (C) RRI and CD45brightCD3neg cells, and (D) EF and CD45brightCD3neg cells. Number of samples = 39; *significant at P < 0.05; **significant at P < 0.01; r = Spearman's rank correlation coefficient.
Figure 7
 
CD45brightCD3pos and CD45brightCDneg correlated to RRI and EF. Correlation between (A) RRI and T cells, (B) EF and T cells, (C) RRI and CD45brightCD3neg cells, and (D) EF and CD45brightCD3neg cells. Number of samples = 39; *significant at P < 0.05; **significant at P < 0.01; r = Spearman's rank correlation coefficient.
Table
 
Demographic Features
Table
 
Demographic Features
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×