Optical Density Ratio of Subretinal Fluid in Choroidal Melanomas Versus Choroidal Naevi Assessed by Optical Coherence Tomography

Abstract

Purpose: The purpose of this study was to determine whether optical density ratio (ODR) of subretinal fluid (SRF) on optical coherence tomography (OCT) differs between choroidal naevi and melanomas.

Methods: One hundred ninety-nine patients (one eye per patient) presenting choroidal melanoma or choroidal naevus with SRF on OCT, evaluated between February and June 2019, were retrospectively included. Other retinal conditions, opaque media, and low-quality OCT were excluded. Mean pixel intensity of SRF (range = 0–255) was quantified using a semi-automated procedure by a masked observer on standard horizontal OCT sections. Mean vitreous intensity served as the reference for ODR.

Results: One hundred twenty-eight patients with choroidal melanoma and 71 patients with choroidal naevus were included in this study. ODR (mean ± SD) was higher in melanomas (181 ± 64) than in naevi (78 ± 48, P < 0.0001). ODR was correlated to lesion thickness (P < 0.0001, r = 0.27), largest basal diameter (P = 0.028, r = 0.16) and, among naevi, to the number of risk factors for growth into melanoma (P = 0.032, r = 0.22). Among 110 patients with naevi or melanoma who underwent fluorescein angiography, ODR was 120.7 ± 550.1 in eyes presenting angiographic pinpoints versus 14.19 ± 26.0 in eyes that did not (P = 0.06). Fourteen eyes with naevi that transformed into melanoma over 3 years had a mean baseline ODR of 94.7 ± 243.5 compared to 4.01 ± 9.74 in 28 matched naevi eyes of similar size that did not transform (P = 0.027).

Conclusions: SRF ODR is higher in choroidal melanoma compared to choroidal naevi. This OCT-derived imaging marker is also higher in choroidal naevi with the potential to transform into melanoma, compared to stationary naevi.

Uveal melanoma is a life-threatening intraocular malignancy, most often developing in Caucasian individuals, involving primarily the choroid and less frequently the ciliary body or iris.^1,2 

Uveal melanoma has a high tendency to metastasize, resulting in a 25-year mortality rate of 50% or more.^3,4 Metastases occur most often to the liver, lungs, and bones, and the median survival after metastasis is 6 to 12 months.⁵ Long-term survival after diagnosis of metastatic disease is uncommon, with only one percent of patients surviving after 5 years.⁵ 

Shields et al.⁶ have produced a comprehensive analysis of the metastatic risk in 8033 patients with uveal melanoma according to tumor thickness, and found that each one-millimeter increment increases the risk of metastasis by 5%. 

Therefore, to detect choroidal melanoma at the earliest possible stage has a paramount importance to reduce metastasis-related mortality. The TFSOM-DIM system used routinely in the clinical setting (thickness > 2 mm, presence of subretinal fluid (SRF), presence of orange pigment, best-corrected visual acuity of 20/50 or less, hollowness on ultrasonography, and largest basal diameter > 5 mm) is to date one of the most widely used to identify suspicious naevi and small melanomas,^7,8 and illustrates the need for objective and reliable imaging markers. 

Advances in high-resolution optical coherence tomography (OCT) imaging allow to capture images of the retina, vitreous, and subretinal space using their infrared reflectivity, at a definition superior to human vision discrimination. Among acquired quantitative parameters, the pixel reflectivity value, on a 256-level grayscale, may reflect the biological properties of ocular media, and could serve as a biomarker. Because SRF is frequently associated to suspicious choroidal naevi, and even more so to choroidal melanoma, and is already considered a risk factor in the TFSOM-DIM system for transformation of benign into malignant lesions, we wondered whether SRF optical density could provide additional information on the transformation risk factor. 

To date, few studies have evaluated the optical density of SRF using OCT, only two in the context of intraocular tumors, but not comparing choroidal naevi and melanomas.^9,10 It has been shown that optical density measurements of SRF differ between degenerative and exudative macular disease,^11,12 and can be used as a biomarker to differentiate various etiologies of SRF.^11,13 

The purpose of this study was to determine whether optical density ratio (ODR) of SRF on OCT differs between choroidal naevi and choroidal melanomas. 

Methods

Patient Selection and Data Collection

Consecutive patients evaluated between February and June 2019 for choroidal naevus or treatment-naive choroidal melanoma who underwent OCT imaging and presented perilesional SRF were retrospectively included in this study. This study adhered to the tenets of the Declaration of Helsinki, and was approved by the Ethics Committee (Internal Review Board) of the French Society of Ophthalmology. The need for written consent was waived due to the retrospective design. 

Exclusion criteria were amelanotic lesions (due to the possible misdiagnosis with other amelanotic choroidal tumors), lesions without SRF, patients co-presenting other retinal conditions (age-related macular degeneration, central serous chorioretinopathy, diabetic retinopathy, etc.), and those without available OCT images. 

Medical records were reviewed to retrieve the following data at diagnosis: demographic characteristics (e.g. sex and age), best-corrected visual acuity, tumor thickness and largest basal diameter, TNM classification, distance to fovea, presence of orange pigment, ultrasound hollowness (corresponding to the TFSOM-DIM criteria), and pinpoints on fluorescein angiography. Melanoma diagnosis and characteristics of naevi, such as ultrasound hollowness, orange pigment, or presence of angiographic pinpoints, were based on observations by attending senior ocular tumor specialists at our department (authors N.C., C.L., L.L., or A.M.). The tumors largest basal diameter and thickness were assessed on ocular B-mode ultrasonography using the ABSolu device (Qantel Medical, Cournon d'Auvergne, France). The shortest distance from the lesion to the fovea was assessed based on color fundus photographs by a single masked observer (author Z.M.). 

In June 2022, a search was performed on our updated patient database to identify possible transformation of choroidal naevi into choroidal melanomas over the 3 years following the inclusion period. 

Optical Coherence Tomography Acquisition

All OCT examinations were performed at our institution on a single device (Angiovue RTx100; Optovue Inc., Fremont, CA, USA). In patients with choroidal melanoma, the OCT at diagnosis was analyzed, whereas in choroidal naevi, the OCT acquired at the selected visit was analyzed. The single perilesional cross-section OCT B-scan with the highest image quality score and containing the largest area of SRF was chosen. The OCT acquisition software scores the quality of each image based on the signal-to-noise ratio, providing an objective and reproducible image selection criterium. The OCT scans were exported from the OCT acquisition software in the compression-free JPEG format, preserving image resolution. 

Horizontal OCT sections at the level of the macula were evaluated by a single observer masked to the diagnosis, to assess the presence or absence of posterior vitreous detachment. 

Optical Density Ratio Measurements

ODRs measurements were performed using ImageJ software (ImageJ 1.53 with java 1.8.0_172, http://imagej.nih.gov/ij/, provided in the public domain by the National Institutes of Health, Bethesda, MD, USA). Image processing and measures were performed on anonymized images by an independent investigator (author Z.M.) blinded to the diagnosis of choroidal melanoma or naevi. Two regions of interest (ROI) were defined on each OCT, one for the SRF compartment and one for the vitreous space (Fig. 1). ROIs were defined along the same vertical axis (corresponding to the anteroposterior axis of the eyes), to limit artefactual fluctuations associated with refraction heterogeneities of intraocular structures (corneal opacifications, cataract, vitreous floaters, refractive aberrations, or other causes of OCT infrared signal modifications in the posterior segment). The widest possible quadrangular-shaped ROI, among four standardized sizes, was defined to assess SRF optical density, according to the subretinal space morphology (30 × 30 pixels, 30 × 15 pixels, 20 × 10 pixels, or 15 × 2 pixels; see Fig. 1). In the vitreous, a standardized quadrangular-shaped, 50 × 50-pixel ROI was defined. ROIs were chosen in a manner that avoids tissue–fluid interfaces or any hyper/hyporeflective debris or other artifacts that might influence the measurement. 

Figure 1.

 

Standardized image processing method for measurement of optical density in subretinal fluid (SRF) (A) and vitreous (B), on optical coherence tomography (OCT). The ImageJ software was used to define regions of interest (C), quantify the average pixel intensity over these areas (D) and provide optical densities (E). Optical density ratios (ODRs) were obtained by dividing the mean intensity of SRF by the mean intensity of vitreous.

Mean optical densities of ROIs were computed as the average of gray level intensity of all pixels (from 0 to 255) inside the ROI. Normalized ODRs with the vitreous used as reference were computed according to the following formula:  

\begin{eqnarray*} \textit{Optical density ratio} &=& \textit{Optical density } [ \text{SRF}]/\\ && \textit{Optical density }[ {{\rm{Vitreous}}}]. \end{eqnarray*}

 

The use of a normalized ratio attempts to neutralize artifacts due to the properties of the image itself (such as picture angle, quality, media opacity, and eye refraction). 

Statistical Analyses

Descriptive and comparative statistics were performed on GraphPad Prism (version 9.4.0; GraphPad Software, La Jolla, CA, USA). The Mann-Whitney test was used for comparisons of quantitative continuous values, and the Chi-square or Fisher's exact test were used for contingency analyses, where appropriate. The data were explored for possible correlations using the Spearman correlation coefficient. Matched comparisons were carried out using the “full optimal” strategy of the “MatchIt” Packlage on R software (version 4.2.3; R Foundation for Statistical Computing, Vienna, Austria, 2021). The volume of naevi was approximated as a sphere section using the formula, volume = (1/3)πh²(3D/2 − h), where, height h is the naevus thickness, and D is the largest basal diameter of the naevus assessed on B-mode ultrasonography. Any P values of 0.05 or less were considered significant. 

Results

Between February and June 2019, of 1481 patients evaluated for fundus lesions, 128 patients with choroidal melanoma and 71 patients with choroidal naevus, who harbored SRF on OCT, were retrospectively included. Their demographic, clinical, and imaging characteristics are reported in Tables 1 and 2. There was no difference in age (P = 0.42), gender (P = 0.66), or tumor distance to the fovea (P = 0.64) between patients with choroidal naevi and melanomas. Consistently, choroidal melanomas had higher lesion thickness (P < 0.0001) and largest basal diameter (P < 0.0001) than naevi. Among the subset of 110 patients who underwent fluorescein angiography, melanomas were more frequently associated with angiographic pinpoints than naevi, although this difference was nonsignificant (P = 0.09). 

Table 1.

 

Descriptive Characteristics of 71 Choroidal Naevi and 128 Choroidal Melanomas With Subretinal Fluid on Optical Coherence Tomography

 

Table 2.

 

Imaging Characteristics of 71 Choroidal Naevi That Presented Subretinal Fluid on Optical Coherence Tomography at Baseline, With and Without Documented Growth Over a 3-Year Follow-Up

 

The mean SRF ODR (± standard deviation) with vitreous intensity as reference, was 181 ± 722 for melanomas versus 78 ± 402 for choroidal naevi (P < 0.0001; Table 3, Fig. 2). 

Table 3.

 

Subretinal Fluid Optical Density Ratio of 128 Choroidal Melanomas and 71 Choroidal Naevi, With Reflectivity of Vitreous as Reference

 

Figure 2.

 

SRF ODR in 128 choroidal melanomas and 71 choroidal naevi, with vitreous reflectivity as the reference.

To assess whether posterior vitreous detachment had an influence on vitreous optical density and SRF ODR, we assessed the vitreous status on OCT images. The vitreous was posteriorly detached in 135 eyes (90 of 128 melanomas, 70%; and 45 of 71 naevi, 66%; see Table 1). Overall, there was no difference in vitreous optical density (7.53 ± 11.2 vs. 7.93 ± 11.6, P = 0.82) and SRF ODR (110.8 ± 472.2 vs. 164.4 ± 780.0, P = 0.55) between eyes with and without posterior vitreous detachment. 

Over a 3-year follow-up, between 2019 and 2022, there were 14 cases of transformation of choroidal naevi into melanoma that were recorded among the 71 naevi with SRF initially included. Given the greater proportion of stationary naevi in the study cohort, we ran a matched analysis with a 2:1 ratio, in order to compare 14 transforming naevi and 28 matched stationary naevi of similar size. The matched populations had comparable distributions with mean baseline naevus volume (± SD) of 4.48 ± 4.53 mm³ and 4.46 ± 4.92 mm³, respectively (P = 0.49). ODR was significantly higher among transforming than stationary naevi (94.71 ± 243.5 vs. 4.01 ± 9.74, P = 0.027; Table 4). 

Table 4.

 

Subretinal Fluid Optical Density Ratio of 14 Choroidal Melanomas That Transformed Into Melanoma Over a 3-Year Follow-Up, and 28 Size-Matched Choroidal Naevi That Did Not Transform, With Reflectivity of Vitreous as the Reference

 

In a correlation analysis of all lesions (Table 5), SRF ODR was correlated to lesion thickness (P < 0.0001, r = 0.29) and largest basal diameter (P = 0.019, r = 0.15) but not patient age (P = 0.21). Among choroidal naevi only, when considering the number of risk factors for transformation into melanoma according to the TFSOM-DIM system (thickness > 2 mm, presence of SRF, presence of orange pigment, best-corrected visual acuity of 20/50 or less, hollowness on ultrasonography, and largest basal diameter > 5 mm),^7,8 there was a positive correlation between the number of factors and the ODR (P = 0.032, r = 0.22). 

Table 5.

 

Correlation Matrix of Clinical and Imaging Characteristics to Subretinal Fluid Optical Density Ratio

 

When considering choroidal naevi and melanoma altogether, 110 patients underwent fluorescein angiography, among which 83 patients presented angiographic pinpoints. The SRF ODR of patients with pinpoints was 120.7 ± 550.1, compared to 14.19 ± 26.0 in those without pinpoints (P = 0.060). 

Discussion

In the present study, we observed that ODR of SRF was significantly higher in choroidal melanomas than in choroidal naevi. Moreover, a range of additional findings supported this observation, including a significantly higher SRF ODR among choroidal naevi that later transformed into melanoma over a 3-year follow-up, compared to a subset of matched choroidal naevi of similar size that did not transform. ODR of SRF had been previously explored in choroidal tumors,^9,10 but, to the best of our knowledge, had not been comparatively analyzed in choroidal melanomas and naevi. 

These results may suggest that the composition of SRF differs in choroidal naevi and melanoma. Although OCT cannot reveal the molecular composition of SRF, which would require spectroscopic investigations, not currently available in commercial devices, one can assume that a higher reflectivity in the infrared spectrum results from a greater cellular or molecular density.¹⁴ Sonoda et al.¹⁵ have explored the effect of blood components on OCT reflectivity, and concluded that it is most strongly affected by the concentration of lipids and proteins. Consistently, the SRF in malignant choroidal melanoma has usually a slightly yellowish color visible on fundoscopy, historically described as “albuminous.” The molecular composition of SRF in choroidal melanoma has been seldom described, due to the complexity and risks of SRF sampling in the context of an active malignant tumor, but the presence of certain proteins at higher concentration than in the vitreous has been ascertained, for instance, the S-100 protein.¹⁶ We can therefore presume, in agreement with other groups, that the diffuse SRF reflectivity in choroidal melanoma originates rather from higher molecular concentration (most likely proteins and/or lipids) than from an abundant presence of cells.¹⁵ Moreover, in posterior uveitis^17,18 or vitreoretinal lymphoma,¹⁹ invasion of the subretinal compartment by cells results in well-defined hyper-reflective dots and/or subretinal hyper-reflective infiltrates on OCT. These imaging signs are less frequent and less intense in melanocytic lesions. 

Two mechanisms may explain the presumed higher protein concentration of SRF in choroidal melanoma. First, extravasation from immature tumor vessels, recently formed by neo-angiogenesis and harboring fragile endothelial barrier, is most frequent in malignant tumors. Choroidal melanoma is known to secrete VEGF and other vasoactive factors stimulating the growth and increasing the permeability of blood vessels.^20–22 The paradoxical uncoupling of endothelial junctions by VEGF, but also Angptl4, CCL2, among other factors, has since long been identified as a pan-cancer mechanism, facilitating tumor cell extravasation and metastatic spread.^23,24 Second, leakage probably also occurs through an altered outer blood retinal barrier, formed by retinal pigment epithelium (RPE) intercellular junction, also disrupted by VEGF and other vasoactive factors,^25,26 and exacerbated in the context of tumor-induced inflammation.²⁷ This outer blood-retinal barrier rupture manifests angiographically by pinpoints, a canonical feature in suspicious choroidal naevi and melanomas, which consist in uni- or plurifocal leakage from the choroidal circulation into the subretinal space.^28,29 Furthermore, long-standing exudative retinal detachments are probably associated to an accumulation of leaked molecules. Thus, the higher ODR in melanomas compared to naevi might also be exacerbated by the longer duration of retinal detachment. However, the influence of SRF duration remains hypothetical because one cannot determine retrospectively when an intraocular tumor and the associated detachment have developed. Underlying pathophysiological mechanisms should be explored in future studies using a translational approach combining multimodal imaging and biological explorations of ocular fluid samples, if accessible. 

Over recent years, advances in high-resolution OCT has allowed to fine-tune the understanding of vitreoretinal disorders, by developing new tools based on imaging routinely performed in clinical setting, such as en face OCT³⁰ retinal layer segmentation,³¹ volume-rendered tridimensional OCT,³² correlation to angiography using confocal OCT,³³ OCT angiography,³⁴ and others. Few previous reports have explored ODR, an OCT-derived quantitative parameter, to analyze SRF using OCT, suggesting that it could be used as a biomarker to differentiate between various etiologies of SRF. This technique relies on accessible image processing methods, using a publicly available quantitative imaging software, and yields standardized ratios by using either the vitreous, the RPE or less frequently the retinal nerve fiber layer (RNFL) as reference. Baek et al.¹³ reported that the ODR of SRF was higher in patients with polypoidal choroidal vasculopathy than in those with chronic central serous chorioretinopathy. Leshno et al.³⁵ observed an increase over time of SRF ODR in rhegmatogenous retinal detachment. Neudorfer et al.¹¹ found that the SRF ODR was higher in patients with vascular retinopathy (age-related macular degeneration, diabetic retinopathy, central serous retinopathy, and pseudophakic cystoid macular edema) than in those with rhegmatogenous retinal detachment or retinoschisis. Finally, the two studies mentioned above have investigated the SRF ODR in choroidal tumors. Leshno et al.⁹ reported a significantly higher SRF ODR in 25 patients with choroidal melanoma than in 14 patients with choroidal metastasis of similar age. They computed using a receiver operating characteristic (ROC)-curve approach with a cutoff value of 0.771 with sensitivity of 78.6% and specificity of 72.0%. Another study by the same group by Zur et al.¹⁰ also identified a higher SRF ODR in 25 cases with choroidal melanoma, compared with 34 cases with circumscribed choroidal hemangioma. 

Our results are consistent with the observations by Leshno et al.⁹ and Zur et al.^9,10 regarding the higher ODR of SRF in choroidal melanoma than in other benign or malignant choroidal lesions. Our findings demonstrate that choroidal melanoma have also higher SRF ODR than choroidal naevus. To date, no comparisons had been performed between choroidal melanomas and naevi. Although our findings are interesting from a pathophysiological and multimodal imaging point of view, there is an important overlap between melanomas and naevi ODR values, which does not allow to use ODR in its current form as a diagnostic tool to discriminate naevi from melanomas. 

Currently, the TFSOM-DIM system elaborated and recently revised by Shields et al.⁷ is used in routine practice to identify suspicious naevi at risk of transformation into melanoma. Interestingly, we observed that the number of factors harbored by each of the 71 naevi analyzed in our study was correlated to the SRF ODR, indicating that this biomarker provides the relevant information consistent with the TFSOM-DIM system. SRF ODR, lesion thickness, largest basal diameter, and the number of TFSOM-DIM risk factors for growth were all correlated (Table 5), a finding consistent with the higher ODR observed in melanomas, and suggesting that thickness and largest basal diameter probably remain the major parameters influencing the level of ODR. Futures studies with a larger cohort could evaluate whether the ODR proves useful as an additional risk factor for growth.^7,36 Artificial intelligence-assisted investigations could prove also very helpful in specifying the relevance of this factor, as suggesting on a recent “white paper” recapitulating past, current, and future tools to discriminate naevi at risk.³⁷ 

This study describing a potential imaging biomarker had several limitations, including its retrospective, mono-institutional design. Although OCT imaging is performed routinely in our center according to a standard acquisition protocol, we had no systematic control over all acquisition procedures, which were performed by residents and orthoptists during patient visits. Furthermore, the technique for defining ROIs and measuring ODRs was also standardized using objective rules, but included semi-automated processes, such as the selection of the largest possible ROI size and exact location within the vitreous and subretinal space. The presence or absence of posterior vitreous detachment on OCT, did not influence the vitreous optical density. Several studies have previously demonstrated the usefulness of OCT to assess the vitreous status,^38–40 including one study using OCT-derived optical density that investigated the anterior vitreous.⁴¹ 

As discussed above, other structures than the vitreous, for instance, the RPE, can be selected as the optical density reference. Explorative analyses (data not shown) had demonstrated that using either vitreous or RPE as reference produced very similar results, and for clarity and homogeneity a single reference (vitreous) was selected. Finally, certain analyses were carried out on a subset of patients, such as statistics on angiographic pinpoints because fluorescein angiography had not been performed on all patients, and certain statistical comparisons demonstrated tendency but remained nonsignificant, due to weak statistical associations or to the limited size of the study population. 

To conclude, SRF ODR is a potential quantitative biomarker that reflects biological changes associated with choroidal naevi and melanoma. Further basic and translational studies on larger populations will be needed to assess its exact pathophysiological meaning and clinical usefulness. 

Acknowledgments

The authors are grateful to Francine Behar-Cohen, MD, PhD, and Thibaud Mathis, MD, PhD, for fruitful discussions regarding the manuscript. 

Disclosure: Z. Mahoun, None; D. Malaise, None; L. Lumbroso-Le Rouic, None; C. Levy-Gabriel, None; N. Cassoux, None; A. Matet, None