Influence of retinal eccentricity, color blindness, and age on optoretinography measured with AOSLO

Abstract

Optoretinography (ORG) is a recent technique for assessing photoreceptor function by measuring their physiological responses to a flash of light. These responses induce changes in the optical properties of photoreceptors, which can be analyzed to evaluate cone photoreceptor health. Recent studies suggest that ORG could be a useful biomarker for detecting retinal pathologies. However, the ORG signal depends on various non-pathology-related factors that need to be taken into account for effective clinical translation. Here, we introduce a new ORG metric and mapping based on the percentage of cones responsive to the stimulus, and we study the effects of retinal eccentricity, color blindness, and age on intensity-based ORG (iORG) using an adaptive optics scanning laser ophthalmoscope (AOSLO).

1. Introduction

Current clinical methods used to assess photoreceptor function can essentially be grouped into two categories: i) subjective, including visual acuity and perimetry measurements, where the patient responds as to whether he sees a stimulus; and ii) objective, where a retinal response is measured directly from the tissue, and which up until recently was limited to electroretinography (ERG). In a typical ERG, either a contact lens embedded with an electrode or a fiber electrode is placed on the patient’s cornea, a flash of light is projected onto the retina, and the resulting electrical response is recorded. However, the captured signal reflects not only the activity of the stimulated photoreceptors but also that of other retinal cells, such as ganglion cells. As a result, the ERG signal represents a complex summation of multiple cellular responses, making it challenging to interpret [1]. So, even though this technique is widely used to evaluate the impact of pathologies on retinal function, its lack of sensitivity and resolution prevents it from assessing photoreceptor function on an individual scale.

On the other hand, progress in imaging devices such as Scanning Light Ophthalmoscopy (SLO) and Optical Coherence Tomography (OCT) has led to significant improvements in sensitivity and resolution. In particular, adding adaptive optics (AO) to classic imaging devices has allowed correction for ocular aberrations, enabling the resolution of individual photoreceptors [2]. Consequently, photoreceptor structure has been largely studied. However, in some pathological cases, photoreceptor structure appears intact despite loss of function or conversely, areas have measurable function despite apparent absence of photoreceptor structure [3,4]. Thus, there is a need for a new tool that would objectively and non-invasively assess photoreceptor function at a cellular level.

Optoretinography (ORG) is a recent technique that relies on photoreceptors’ physiological response to a visible light stimulus [5,6]. When photoreceptors are stimulated with visible light, a response can be extracted from en face images which corresponds to intensity scintillations of the cones. This is believed to be due to changes in the optical path length of their outer segment [7] leading to a change in interference between light reflected by the inner/outer segment junction (IS/OS) and the cone outer segment tips (COST) layers. With AO imaging devices, videos of the photoreceptor layer are recorded during which a flash of visible light is sent to the retina. Photoreceptors’ reaction to this stimulus is then analyzed to extract an ORG signal enabling us to determine whether they are functioning. Therefore, ORG has the potential to become a new biomarker for the study or monitoring of retinal pathologies, with the added advantage that it can be measured with different imaging devices. While ORG recorded with OCT, which measures optical path length changes in the outer segments, are called phase ORG (pORG), ORG recorded with en face AO ophthalmoscopes are based on changes detected in the intensity of photoreceptors and are called intensity ORG (iORG) [8].

Intensity-based ORG were first recorded by Jonnal et al [7] with a flood illumination AO device and Grieve et al [9] with an AOSLO. Then, Cooper et al [10] studied the influence of the stimulus wavelength and intensity on the cone photoreceptor’s intrinsic response, establishing the ORG as a direct measure of photoreceptor function. A few years later, they were able to resolve the intensity ORG at the level of individual cones and to separate the population of S cones from the population of L and M cones [8]. Repeatability and reciprocity of iORG signals were then studied by Warner et al [11] who found a good repeatability between different imaging sessions on the same subject but also found differences between subjects. This showed that iORG seems to be a robust tool, but is sensitive to many factors that have to be taken into account when analysing the result.

Ourselves [12] and Warner et al [13] have noted differences related to retinal location. Another factor of importance is the response of cones to different stimulus wavelengths. Particularly, the response to a stimulus with a specific visible wavelength is different for the three types of cones: short (S), medium (M), and long (L). Consequently, color blindness has an important repercussion on the ORG signal, depending on the stimulus wavelength and the type of color blindness. Zhang et al [14] studied one type of color blindness: deuteranopia (no green cones). They performed phase-based ORG on one subject using an AO-OCT and three different stimulus wavelengths (637nm, 528nm, 450nm), allowing them to differentiate and map the different types of cones. As expected, the deuteranope subject presented no green cones and thus, an altered ORG signal whereas he was free of ocular disease and had best corrected visual acuity of 20/20. Finally, an important parameter influencing ORG signal that has not yet been tackled in the ORG literature is age. When moving to clinical application, the most appropriate age-matched controls might be more elderly individuals if considering age-related pathology.

Consequently, to be able to properly interpret the ORG signal, we need a baseline that takes into account the variability in healthy subjects. This is particularly important if we want to use ORG as a biomarker of pathologies, which has already been done on patients having retinitis pigmentosa [15,16] and choroideremia [17]. In these papers, the authors found differences between normal subjects and patients with retinal pathologies.

In this paper, we introduce a new metric based on the proportion of cones that respond strongly to stimulation during each acquisition, combined with a cone mapping of the distribution of these responsive cones for easy visualization of abnormally behaving zones. Additionally, we examine the impact of retinal eccentricity and color blindness on iORGs measured with AOSLO. Then, we compute iORGs on a larger healthy aging cohort to study the influence of age.

2. Methods

2.1. Imaging device

We recorded ORG signals using an AOSLO (a custom modified MAORI, Physical Sciences, Inc., Andover, MA, USA), which has been previously described [18]. Briefly, the imaging source was a superluminescent diode Exalos SLD centered at 757 nm with a 20 nm full-width at half maximum, leading to a coherence length of 8.8 μm. This source gave no detectable ORG but was used to get structural information, as it gave speckle-free images. In order to record ORG, we added a new source to the existing one with an electronic switch allowing to pass from one source to the other. To record the optoretinograms, the source we used was a diode LP-730-SF15 centered at 727 nm with a full width at half maximum of 2 nm, leading to a coherence length of 82 μm. This long coherence length was necessary to record the strongest ORG signal according to the literature [7] but it also led to images containing speckle. Two galvanometer scanners were used to raster scan the retina over a field of view of 1^∘ by 1^∘ at an imaging rate of 37.5 Hz. Simultaneously, an adaptive optics corrected OCT B-scan was acquired with a 840 nm beam, which served as an orthogonal live OCT view of the retina and as the wavefront sensor beacon. To correct for ocular aberrations, an adaptive optics loop was employed, consisting of a Shack-Hartmann wavefront sensor and a deformable mirror (DM69, Alpao, France), operating in a closed loop at a frequency of 10 Hz.

The AOSLO system optical layout was modified to add a stimulus path allowing a visible flash of light to be delivered to the photoreceptor layer during imaging. A single-mode fiber coupled to a laser diode (QPhotonics QFLD-520-10S, 520 nm, 450 μW at the cornea) was combined with an acousto-optic modulator (AOM) which allowed us to gate the stimulus delivery and modulate its shape [19]. The stimulus beam followed the same optical path as the imaging light to ensure that the stimulated area was the same as the imaged one. A fixation target was used to guide the subject’s gaze, allowing us to choose the retinal area to image.

2.2. Imaging protocol and cohorts

The study was conducted in accordance with the Declaration of Helsinki, and approved by the Ethics Committee in France (IDRCB number: 2021-A00410-41, CPP Est I). Imaging sessions were held on healthy volunteers who provided informed consent prior to inclusion. In total, 52 people were imaged (among whom 24 were female) with an average age of 46 years old.

Each subjects’ pupil was dilated and accommodation arrested using one drop of tropicamide (Mydriaticum 2mg/0.4mL). We started the imaging session approximately 30 minutes after the drop, when accommodation should be maximally blocked for a one-hour time interval. Each acquisition consisted of a 5s-image sequence: 2s of pre-stimulus, then the green stimulus was delivered to the field of view during 1s and then 2s of post-stimulus were recorded. In total, each acquisition was made up of 185 frames (around 5s). We performed two preliminary studies to determine important parameters of the imaging protocol: the time of dark-adaptation and the recording duration. Details of the protocols and results of these studies are developed in Supplement 1 ^{(2MB, pdf)} (Section 1).

2.3. Extracting the intensity optoretinogram

Image sequences were first deinterlaced to remove the sinusoidal artifacts created by the movements of the scanners. Then, videos were registered using a strip-based registration method [20] which corrected for the natural eye movements. After that, each registered frame was used to compute an averaged frame of the whole video. On this averaged image, cone photoreceptors were detected using a customized ImageJ plugin based on Gaussian difference, and the coordinates of each cone were saved.

Then, we extracted the ORG signal using the approach described previously by Cooper et al [10]. Briefly, we extracted each cone photoreceptor intensity by averaging a circular patch of radius 3 pixels around each cone centroid, for each frame. By doing so, we obtained a graph representing each cone’s intensity as a function of time. At this stage, we removed cones considered as outliers because of their really high or really low intensity. We eliminated cones of intensity exceeding $I_{outliers}=m\pm 2\times \sigma$ where m and $\sigma$ are the mean and standard deviation averaged over all cones at each time point of the acquisition. At this step, when considering all subjects, an average of 3% $(\pm 1\mathrm{\%})$ of cones were removed. Then, cones whose pre-stimulus signal was detected in less than 60% of pre-stimulus frames (e.g. due to blinks, movements, etc) were also excluded from analysis, as in Cooper et al [8]. At this stage, an average of 24% $(\pm 11\mathrm{\%})$ of remaining cones were excluded from analysis. Note that these exclusions follow the literature and are different from the percentage of responsive cones defined later. Then, for each frame, we divided each cone intensity by the mean of all cone intensities, allowing us to get rid of frame-wide intensity changes. Each cone’s post-stimulus intensity (from the first frame stimulated to the last frame of the recording) was then standardized by its pre-stimulus intensity. We then computed $S(t)$ , the standard deviation at each time point, over all cones, allowing us to measure the dispersion of cone responses after the stimulus according Eq. (1).

$$S(t)=\sqrt{\frac{1}{n-1}\sum _{i=1}^n{|R_i(t)-m(t)|}^2}$$ (1)

where n is the total number of cones, $R_i(t)$ represents the standardized intensity of cone i at time t, $m(t)$ is the mean intensity of all cones at time t. Standard deviations from all stimulus acquisitions on the one hand, and all control acquisitions on the other, were then pooled to obtain an average stimulus curve and an average control curve. The final ORG curve was then obtained by taking the difference between the stimulus and the control curves, and a smooth spline was fitted to this final curve (MathWorks, Natick, MA, USA, R2024b, function spaps with a smoothing parameter 0.995). For each subject, the peak of this smooth spline was then extracted, and called the ORG amplitude. The slope of the rising portion of the curve was determined by computing the derivative of the fitted spline and identifying its maximum value.

2.4. Thresholds

We then introduced a new metric to detect those cones which respond the most strongly to the stimulus, as described in Fig. 1. To achieve this, we needed to define thresholds that would divide cones into responsive or non-responsive categories.

Fig. 1.

Overview of population thresholds calculation and use. A) Concatenation of the intensity curve of each cone of all control acquisitions. Each curve corresponds to the normalized intensity of a control cone. B) Histogram of post-stimulus means. Thresholds are computed on this distribution (mean +/- std) and are represented by the two black lines. On this example, 10% of the cones are outside the thresholds. C) Concatenation of the intensity curve of each cone of all stimulus acquisitions. Each trace is colored according the behavior of the cone intensity after the stimulus : green for cones whose intensity increases more than the upper threshold, purple for cones whose intensity decreases more than the lower threshold, and orange for cones whose intensity stays between the two thresholds. D) Histogram of post-stimulus means. Black lines represent the population thresholds computed in B. On this example, 24% of the cones are outside the thresholds and are therefore considered to be responsive.

To do so, on one healthy subject (female, 26 years old), we took 7 stimulus acquisitions and 3 control acquisitions at 3^∘ temporal where the subject maintained a good fixation. We were then able to accurately register these acquisitions, i.e. to single cone scale over multiple movies, and crop them to their common area. We then performed cone detection as described before. After these steps, we were able to follow each cone photoreceptor individually over the course of these 10 acquisitions. We performed the post-processing steps explained in the paragraph above until obtaining the standardized intensities. We then defined two different thresholding approaches:

• 1.Population threshold: On the three control acquisitions, the mean (m) and the standard deviation $(\sigma )$ were computed on the frames that correspond to the post-stimulus period, taking into account all analysed cones. Then, the upper and lower thresholds were defined as follows: $T_{+}=m+\sigma$ and $T_{-}=m-\sigma$ .
• 2.Individual threshold: For each photoreceptor, we computed its upper and lower thresholds based on its individual normalized intensity from the three control acquisitions. Consequently, for each cone, two thresholds were defined as follows: $T_{i+}=m_i+{\sigma }_i$ and $T_{i-}=m_i-{\sigma }_i$ , where $m_i$ and ${\sigma }_i$ are the mean and standard deviation of the cone i, computed on the frames corresponding to the post-stimulus.

For each approach, we retained only those cones which exceeded the thresholds and we called them "responsive cones". For both methods, we then analyzed each cone individually to determine how many times it responded across the seven stimulus acquisitions. We were then able to map these cones on the average image where each cone was colored according to the number of times it responded. We plotted the cumulative percentage of cones that had responded across successive acquisitions, in order to estimate how many acquisitions were needed for all cones to respond as expected (i.e. all L and M cones, since S cones are not expected to respond to the green stimulus).

Based on the results of the study presented in the previous paragraph, we chose to use the population threshold method for all the other studies presented in this paper. Figure 1 provides an overview of the method used to extract and apply population thresholds in order to calculate the average number of responsive cones.

We recorded, for each subject, the average percentage of cones that exceeded the thresholds. It should be noted that the average percentage of responsive cones differs from the cumulative percentage described above. The average percentage refers to the proportion of cones responding to the stimulus during a single acquisition, averaged across several acquisitions, whereas the cumulative percentage corresponds to the proportion of cones that have responded at least once over the course of multiple acquisitions. We then used these results to perform cone mapping. On the average image of each acquisition, we colored each photoreceptor with a color corresponding to its behavior: green for cones whose intensity increased after the stimulus, orange for non-responsive cones that remained between the two thresholds and pink for cones whose intensity decreased after the stimulus.

2.5. Influence of retinal eccentricity and color blindness on the iORG signal

The aim of these studies was to determine the influence of retinal eccentricity and color blindness on the ORG. Parameters of these studies are gathered in Table 1. As mentioned before, following the results of the analysis described in the previous section (section 2.4) where we studied two different thresholding methods, we used the population thresholding method for these two studies. Reasons for this choice are presented and discussed in sections 3.1 and 4.1.

Table 1. Parameters of the two main studies. F = Female. M = Male. CB = Color Blind. N = Normal.

Investigated parameter	Number of participants	Sex	Age	Eccentricity
Eccentricity	4	4 F	32 (+/-9)	2∘, 3∘, 4∘,5∘ Temporal
Color Blindness	CB: 12 N: 8	CB: 11M N: 4M	CB: 37 (+/-15) N: 32 (+/-7)	2∘ Temporal and Nasal

To study the influence of retinal eccentricity, we recorded ORG on four healthy volunteers (females, 44, 32, 26 and 24 years old) at different eccentricities (2^∘, 3^∘, 4^∘ and 5^∘). For each eccentricity, we took 5 control acquisitions and 5 stimulus acquisitions.

Next, we studied the influence of color blindness on a cohort of twelve color blind volunteers (eleven males, average age 37 years old) and seven healthy volunteers (4 males, average age 32 years old). We recorded 20 image sequences: 10 acquisitions at 2^∘ eccentricity (temporal and nasal) and 10 without any stimulus for control purposes. Because of poor image quality, two of the subjects gave no ORG signal as we were not able to register the frames. The type of color blindness was previously diagnosed using the D15 Lanthony desaturated test [21] and results were computed on the website https://www.torok.info/colorvision/d15.htm. This test determines if the subject has a defiency in their green cones (deuteranomaly), in their red cones (protanomaly) or in their blue cones (tritanomaly). Moreover, an OCT examination was performed (Spectralis, Heidelberg, Germany) to confirm the absence of any detectable retinal pathologies. We then categorized the volunteers based on their diagnoses obtained from the desaturated Lanthony test.

2.6. Study of the influence of age on a healthy aging cohort

To analyze the influence of age, we recorded ORG on a cohort of 32 subjects (among whom 18 females) aged from 22 to 83 years old. Data from six participants were excluded due to poor image quality caused by excessive eye movements, which hindered image registration, and high noise levels that made cone detection impossible. Acquisitions were taken at 2^∘ temporal from the foveal center. The volunteers were then divided into four categories based on their age (see Table 2).

Table 2. Age distribution of the included healthy volunteers for the aging study.

Group	Age limits	Number of people	Average age	Standard deviation
1	20 to 40 yo	12	29.2	4.8
2	40 to 60	4	47.2	3.7
3	60 to 80	7	71.3	5.3
4	80 to 83	3	82	1

To speed up the imaging sessions, we explored a partial field stimulus approach [22]. We chose to deliver the stimulus on half of the field of view only to use the other half as control acquisitions, in order to reduce the imaging session duration, for patient comfort and protocol efficiency. When computing the ORG, cones at the border region (i.e. the central third of the field) between the upper and lower halves were excluded from the analysis to avoid potential interference from fixational eye movements that could lead to unintended stimulation of the control area. Nevertheless, we later observed that the control curves in half-field acquisitions showed an abnormal increase not present in full-field control recordings (see Supplement 1 ^{(2MB, pdf)} Fig. 3), suggesting that light leakage from the stimulus might have affected the control half of the image. As this influence was identified after the study, we concluded that the half-field controls did not represent true control conditions. Therefore, for this cohort, we only used and represented the stimulus curves, which remained consistent between half- and full-field acquisitions.

Fig. 3.

Influence of eccentricity and color blindness. A) and B) ORG signals. Each curve is the average of all subjects of each group. C) and D) ORG amplitude at different eccentricities and for different color blindness conditions. E) and F) Slope of the ORG signal at different eccentricities and for different color blindness conditions. The mean of each group is represented by a white-filled circle with a black outline. Error bars correspond to +/- standard deviation of each group. Results of the one-way ANOVA test are displayed. ns= not significant, * means p < 0.05 and ** means p < 0.01.

2.7. Statistical analysis

For our studies, we wanted to determine if the ORG amplitude, the slope and the percentage of responsive cones differed significantly between the different groups. To do so, we conducted a one-way ANOVA using MATLAB (MathWorks Inc., Natick, MA). Each group consisted of independent measurements of amplitudes, slopes and cone responsiveness. The significance threshold was set at p < 0.05. Following a significant ANOVA result, post-hoc multiple comparisons were performed using Tukey’s Honest Significant Difference (HSD) test to identify which group pairs exhibited statistically significant differences. Group comparisons were annotated with conventional significance levels: p > 0.05 (ns), p < 0.05 (*) and, p < 0.01 (**). For the retinal eccentricity study, on each subject, we attempted to apply a linear regression model to the percentage of responsive cones as a function of retinal eccentricity and we reported the coefficient of determination R².

3. Results

3.1. Comparison between individual and population thresholds

We first studied two different thresholding approaches: population thresholds or individual thresholds. Results of this study are gathered in Fig. 2.

Fig. 2.

Comparison between computing population thresholds (A and C) or individual thresholds (B and D). Top panel displays an image averaged on 10 acquisitions on a healthy volunteer, after cropping to a common area. Colors correspond to how many times each cone was considered responsive on a total of 7 acquisitions with stimulus, when the thresholds were computed on the population A) and individually B). Bottom panel displays the cumulated percentage of responsive cones with population thresholds C) and individual thresholds D). Dotted line indicates 93% of cones, which corresponds to the approximate percentage of cones that we expect to be stimulated since S cones are supposed to be between 5% and 10% [23]. See Visualization 1 ^{(5.5MB, avi)} (population thresholds) and Visualization 2 ^{(5.4MB, avi)} (individual thresholds).

Panels A and B show each cone color-coded according to the number of acquisitions (out of seven) in which it responded to the stimulus. Cones responding across multiple acquisitions are more numerous with the individual than the population threshold. Globally therefore, cones are classed as responsive to more acquisitions when the thresholds are calculated on each individual cone rather than averaged over the population of all cones. Panels C and D show the cumulative percentage of responsive cones at each acquisition, accounting for cones that responded in previous acquisitions. When using population thresholds, after seven acquisitions, 80% of cones have responded to the stimulus at least once whereas this percentage reaches 93% when using the individual method.

3.2. Variation in ORG signals, amplitudes and slopes with eccentricity and color blindness parameters

Figure 3 presents the results of the studies investigating the effects of eccentricity and color blindness. On panel A and B, the ORG curves are displayed, while graphs on panels C, D, E, and F show the influence of these two parameters on the ORG amplitude and slope.

Figure 3(A) shows the influence of retinal eccentricity on the ORG signal. At 2^∘ eccentricity, the ORG curve increases rapidly with the stimulus and then decreases slowly. As we move away from the fovea, not only the highest value reached is lower but also the growth rate is slower (i.e. the slope of the increase is smaller). This behavior is summarized in the graphs presented on panels C and E where we can see a clear decrease of the ORG amplitude and slope with the eccentricity. Nonetheless, with this small number of participants, when computing the one-way ANOVA between the amplitudes or between the slopes, at different eccentricities, none of them was significant.

Regarding the color blindness study, the Lanthony test revealed that, among the ten color-blind volunteers in our cohort, six were diagnosed with deuteranomaly (deficiency on the M-cones), three with protanomaly (deficiency on the L-cones) and one with tritanomaly (deficiency on the S-cones). Figure 3(B), D, and F show the influence of color blindness on the ORG signal, amplitude and slope. A reduced ORG amplitude is observed for the deuteranomalous group (0.4 against 1.1 for normals) whereas an increased amplitude is observed for the protanomalous group (1.3). This observation is supported by the statistical analysis of ORG amplitudes presented in panel D: the difference between the normal and deuteranomalous groups was significant (p = 0.0284, *), whereas no significant difference was found between the normal and protanomalous groups or between the deuteranomalous and the protanomalous groups. Moreover, the slope of the rising part also follows the same trend: it is reduced for the deuteranomalous group whereas it is increased for the protanomalous group. One-way ANOVA test revealed that the difference of slope between normals and deuteranomalous groups was significant (p=0.0012, **) as well as the difference between deuteranomalous and protanomalous group (p=0.0016, **). The difference between the normals and the protanomalous groups was found not significant. The curve of the tritanomalous subject presents high noise and no real detectable ORG signal. We therefore did not consider amplitude and slope on this subject.

3.3. Percentage of responsive cones and cone mapping

Figure 4 shows the evolution of the average percentage of responsive cones for the two studies.

Fig. 4.

Percentages of responsive cones for the two studies. Each point corresponds to one subject. The mean of each group is represented by a white-filled circle with a black outline. Error bars correspond to +/- standard deviation of each group. A) Percentages of responsive cones at 2^∘, 3^∘, 4^∘ and 5^∘ of the four subjects. No significant difference was found between the groups. B) Percentages of responsive cones for the three types of color blindness and the control group. Significant differences (p < 0.05, *) were observed between the normal and deuteranomalous groups and the deuteranomalous and protanomalous groups.

Figure 4(A) shows the influence of retinal eccentricity on the average percentage of responsive cones. The percentage of responsive cones seem to decrease with the eccentricity even if the differences between the groups were found not significant with the one-way ANOVA test. In Supplement 1 ^{(2MB, pdf)} , Fig. 4, we plotted the ORG amplitude as a function of eccentricity and we applied a linear fit for each subject with a coefficient of determination R² superior to 0.95 for two subjects and between 0.2 and 0.5 for the two other ones.

Figure 4(B) shows the influence of color blindness type on the average percentage of responsive cones. The deuteranomalous group presents a lower average percentage than the normal group (12% instead of 22%). The result of the one-way ANOVA revealed a significant difference between the normals group and the deuteranomalous group (p=0.0318, *) and between the protanomalous group and the deuteranomalous group (p=0.0149,*) but no significant difference was found between the normals group and the protanomalous group.

Figure 5 shows two cone mappings, for a healthy volunteer (Fig. 5(A)) and a color blind subject diagnosed with deuteranomaly (Fig. 5(B)). For the healthy volunteer case, 19% of the detected cone photoreceptors are identified as responsive; among these, 12% exhibit increased intensity, while 7% exhibit decreased intensity. In contrast, the deuteranomalous subject presents only 11% of responsive cones (7% increasing and 4% decreasing). In both subjects, the responsive cones appear evenly distributed throughout the field of view.

Fig. 5.

Images obtained by averaging all frames from one acquisition with colored points representing the behavior of each photoreceptor after the stimulus. A) Healthy volunteer, female, 24 years old. B) Color blind subject diagnosed with deuteranomaly, male, 22 years old.

3.4. Larger cohort to study the effect of age

We wanted to study the influence of age on the iORG signal. To do so, we had access to a large cohort of healthy volunteers from 24 to 83 years old. To minimize the imaging session duration, in an effort of improving patient comfort, instead of stimulating the full field of view and then taking other acquisitions without the stimulus to use them as control acquisitions, we chose to stimulate the half field of view only and use the cones located in the other half as the control acquisitions. To verify that the shape of the stimulus was not influencing the results, we performed ORG imaging sessions with both half-field and full-field stimulus on three healthy volunteers. Results of these experiments are available in Supplement 1 ^{(2MB, pdf)} Fig. 3. When displaying the stimulus curves, the match between half field and full field stimulation was good. However, the control curve of the half-field experiment was systematically above the control curve of the full field experiment. We therefore suspect that cones located in the non-stimulated half field of view were actually receiving some stimulation during the stimulus period. As a consequence, the control curves were higher than in "true" control acquisitions. We therefore discarded the control data and in Fig. 6(A) we display only the stimulus curves, without subtraction of controls. Consequently, we computed the stimulus curve amplitudes (Fig. 6(B)) instead of the ORG amplitudes as well as the slopes (Fig. 6(C)) of the rising part of the stimulus curves instead of the ORG curves.

Fig. 6.

Results of the age study. A) Stimulus signals of the four age groups. Note that these curves cannot be compared to the curves of the other studies as we are here displaying the stimulus curves only and not the ORG curves because we could not subtract the control curves as they were found unconclusive. More details in text. B) Amplitudes of the stimulus curves shown in A. C) Slopes of the rising part of the stimulus curves for each age group. In panels B and C, the mean of each group is represented by a white-filled circle with a black outline. Error bars correspond to +/- standard deviation of each group. Results of the one-way ANOVA test are displayed. ns = not signicant, * means p < 0.05, ** means p < 0.01.

Figure 6(A) illustrates the influence of age on the stimulus curves. The youngest age group (20–40 years) exhibits the strongest signal, with a rapid increase and a peak around 3. Middle-aged individuals (40–60 years) show a slightly slower rate of increase and a slightly lower peak, around 2.6. The 60–80 age group also experiences a reduced rate of increase and a peak near 1.8. Finally, the oldest group (80–83 years) presents the slowest increase and the lowest peak, around 1.6. Figure 3(B) displays the stimulus peak amplitude for all subjects. ORG amplitude appears to decrease with age. However, the difference between the youngest group (20–40 years) and the middle-aged group (40–60 years) is not significant (p = 0.8550). In contrast, significant differences are observed between the youngest group and the two older groups, with p = 0.0304 between the first and the second oldest group, and p = 0.0064 between the first and the oldest group (indicated respectively by * and ** in Fig. 6). When looking at panel C, the average slope seems to stays constant between the first two groups and then decreases for the two oldest groups. The only significant difference in slopes between the groups was found between the youngest and the oldest groups (p=0.0069, **).

4. Discussion

We studied the influence of three parameters (retinal eccentricity, color blindness and age) on the ORG signal recorded with an AOSLO. Even if a good repeatability between sessions was already established on healthy people by Warner et al [11], there was a need to study the influence of some parameters that differ between subjects and can influence the ORG signal. We also introduced a new metric: the percentage of responsive cones. To compute it, we categorized stimulated cones into two groups—those showing greater responses in stimulus acquisitions compared to controls, and those with intensity variations similar to controls, indicating no response or a non-measureable response. The classification was based on a threshold defined as one standard deviation around the mean intensity variation in control acquisitions. We empirically tested thresholds of one and two standard deviations and ultimately selected the former, as it yielded a reasonable proportion of responsive cones to be retained for analysis. In contrast, the two-standard-deviation criterion was overly stringent, producing very few responsive cones in healthy subjects and virtually none in individuals with color blind conditions. Then, on a larger cohort, we conducted a study on the influence of age on the ORG signal.

4.1. Individual VS population threshold

In 2020, Cooper et al published their work on optoretinography on individual human cone photoreceptors [8]. They showed that acquisition after acquisition, the same photoreceptor presented highly heterogeneous responses: for some acquisitions, its intensity decreased after the stimulus, for others it increased and for others it oscillated or remained stable. For our part, we found that on healthy subjects, on a single acquisition, the average percentage of cones responding to a green stimulus was approximately 20%. So, following the work of Cooper et al, we wanted to investigate if the same subset of cones consistently responded, or whether different cones were activated across acquisitions. To do so, we tried two different thresholding approaches, allowing us to classify cones into "responsive" and "non responsive" categories. We repeated the same acquisition seven times on one subject and looked at individual cone responses, one acquisition after another. In the first approach, a global population threshold was computed based on the average intensity variations of all cones during three control acquisitions. In the second approach, an individual threshold was calculated for each cone, based on its own particular intensity fluctuations during these three control acquisitions. We found that, regardless of whether a population-based or individual threshold was used, more than 80% of cones responded at least once across the seven cumulated acquisitions. This is notable given that M-cones, which are most sensitive to green light, comprise only about 30–40% of the cone population. These results suggest that L-cones also respond to the green stimulus, which is to be expected due to their overlapping spectral sensitivity with that of M cones [24]. Moreover, the results are also coherent with the percentage of blue cones (5%−10% according to [23]) even in the individual threshold where after 7 acquisitions the total number of responsive cones does not exceed 95%.

The population threshold is calculated from the average control signal across all cones, while the individual threshold reflects the control variability of each cone separately. Therefore, for a cone showing very little variability under control conditions, even a slight deviation upon stimulation will exceed its individual threshold, whereas this small change would remain undetected if a common population threshold were applied. To summarize, our results show that the population thresholding method is less sensitive: cones exhibiting only slight variations following stimulation are not classified as responsive. In contrast, the individual thresholding method yields a higher number of responsive cones, as it accounts for each cone’s intrinsic variability in the absence of stimulus. Each method has its own advantages. The population thresholding approach is simpler to implement, as it does not require alignment of acquisitions; this makes it more robust in routine or clinical conditions where subject movement may occur between acquisitions. The individual thresholding method, however, necessitates precise alignment across acquisitions, requiring the subject to remain stable throughout the imaging session. Despite this constraint, this method offers higher precision, enabling the assessment of cone activity at the single-cell level. Using this approach, we observed that, after only five acquisitions, 92% of cones were activated at least once. These results demonstrate the potential of this method to evaluate individual photoreceptor function in a relatively short acquisition time, which could be particularly valuable for monitoring the effects of targeted interventions such as cell-based therapies. Nevertheless, the population thresholding method remains useful for broader assessments of retinal health. We suggest that a combination of these approaches could be used in clinical application. When the resulting ORG signal amplitude and the average percentage of responsive cones fall within the expected range of healthy individuals, further analysis may not be necessary. However, in cases where deviations from the norm are observed, the individual thresholding method can be employed for a more detailed, cone-level investigation. Moreover, in situations where data are noisier—such as with elderly subjects or certain patient populations—alignment may be challenging.

4.2. Retinal eccentricity

We studied the influence of eccentricity on the ORG signal in a cohort of four healthy volunteers. We found that, as the distance from the fovea increases, the ORG signal decreases in both amplitude and slope. This suggests that cones become not only less reactive but also slower in their response to the stimulus with increasing eccentricity. This result is consistent with findings by Warner et al. [13], who observed a similar trend in a cohort of five healthy subjects across eccentricities ranging from 1^∘ to 16^∘. Jiang et al [25] and Wendel et al [26] also found a significant linear relationship between the eccentricity and the ORG amplitude in healthy subjects, as well as a linear relationship between the outer segment length and the maximum change in optical path length (OPL).

Pandiyan et al [27] attributed the cone outer segment elongation after visible stimulus to an osmotic change caused by the phototransduction. As a consequence, cones with longer outer segments will contain a larger volume and thus will swell over a greater range than those with shorter segments, leading to stronger scintillations in en face images and consequently a larger ORG response. This could explain the decreasing trend of ORG with eccentricity, as the cones outer segment length decreases with eccentricity.

We did not see any effect of age on retinal eccentricity over the three decades we measured here (i.e. subjects aged 24, 26, 32 and 44), but in future, could expand to more elderly subjects.

4.3. Color blindness

We worked with a cohort of 20 participants, including 12 color-blind individuals and 8 control subjects. The color-blind participants were classified into three categories based on their type of color vision deficiency, as determined by the Lanthony desaturated panel test.

AO imaging in dichromats has revealed that individuals with dichromatic vision have either a replacement model of color blindness where one cone type is replaced with another cone type but the individual has near normal cone density or has functional loss of one cone type with gaps in the mosaic and therefore reduced density [28]. In either the replacement or loss models however, the cones in the confocal mosaic will not contain both L and M cones (for protanomalous or deuteranomalous cases). Thus the individuals with functional L-cone loss (protanomalous) have a higher number of M-cones and therefore exhibit a higher percentage of responding cones, as the stimulus at 520nm is stronger for M than L cones. The deuteranomalous individuals have a lower iORG amplitude and slope, as well as a lower percentage of responding cones in comparison to the trichromats and the protanomalous individuals, possibly due to the fact that they may have more L-cones which are slightly less sensitive to the 520 nm stimulus.

The tritanomalous subject did not give a detectable ORG response. In light of this, we retested his color blindness type using the Lanthony test and the result of the second test was "diffuse color discrimination error". We therefore consider that this subject’s color blindness diagnosis is as yet inconclusive and may not in fact correspond to tritanomaly. Interestingly, the unexpected iORG result led us to this further testing. We intend to further explore this interesting case in future work.

4.4. Thresholds and responsive cones

In this study, we introduced a new metric related to the iORG signal: the average percentage of responsive cones. This metric quantifies the proportion of cones in a given acquisition whose signal exceeds the control signal from the same region. One limitation of this threshold method is that in any one particular acquisition, it does not take into account cones whose response to the stimulus was an oscillatory movement instead of a remarkable increase or decrease of intensity. However, as it was shown by Cooper et al [8], a same cone will have different behaviors when repeating stimulus acquisitions. Its intensity may increase on the first acquisition and then decrease in another or oscillate. Consequently, a cone whose intensity may oscillate will be excluded from analysis on those acquisitions where its response is oscillatory, but will then be taken into account when its behavior changes in another acquisition. As we average over several acquisitions, all responsive cones should in the end be analyzed during an imaging session. Nevertheless, it should be noted that we are always underestimating the percentage of responsive cones per acquisition as we do not take into account the oscillatory ones. We also exclude cones for reasons related to fixation stability, as in [8]. Despite this limitation, the percentage of responsive cones may serve as an additional indicator of retinal function that could vary with pathological changes.

We observed a decrease in the percentage of responsive cones with increasing eccentricity. For two out of four subjects, this decrease followed a linear trend (see Supplement 1 ^{(2MB, pdf)} Fig. 4). Perhaps this decrease may be explained by our population thresholding method, which excludes weakly responsive cones. If ORG amplitude decreases at larger eccentricities [27], then our individual thresholding method may be more appropriate to be sensitive to weak ORG changes at larger eccentricities. The percentage of responsive cones metric also proved useful in distinguishing between normal and color blind individuals. In deuteranomalous subjects, the proportion of responsive cones was markedly lower, accompanied by a reduced iORG signal. This observation is consistent with the decreased number of green-sensitive cones in this condition and supports the interpretation that fewer cones are physiologically capable of responding. These findings align with those of Zhang et al. [14], who performed ORG on a deuteranope (a subject completely lacking green-sensitive cones) and mapped cones based on their responses to the stimulus. They identified two groups of cones whose responses matched those of S and L cones in color-normal subjects, while the third group—expected to correspond to M cones—was absent.

We also performed cone mapping by assigning colors to cones according to their behavior after stimulation. This revealed that responsive cones seem to be evenly distributed. In pathological conditions, where cones may be less responsive or absent altogether, the proportion of responsive cones can serve as a measure of functional loss, while cone mapping highlights the spatial distribution of non-responsive cones. This distinction can help determine whether cone dysfunction is diffuse or localized, providing insights into disease mechanisms that go beyond what can be inferred from iORG amplitude alone.

4.5. Age study

Retinal imaging becomes more challenging with advanced age, as participants may blink more frequently and exhibit reduced fixation stability, which complicates frame registration. Moreover, they can also present some opacity in their lens, leading to a weaker signal that prevents the adaptive optics loop from correcting aberrations, resulting in blurred images. The oldest age group (between 80 and 83 years old) was composed of only three people which resulted in a noisier curve. The protocol used for this study was to stimulate only the half field of view, which had the advantage of halving the duration of the imaging session, which was helpful, especially with old people who got easily tired or distracted. However, as we considered that only one third of the field of view was stimulated, the number of analysed cones was smaller, making the result noisier than with a full-field stimulus. In addition, we encountered problems with using a half field control zone, with the control region apparently receiving some stimulation light, thus preventing us from calculating the final iORG curves with control subtracted.

We nevertheless measured a decrease of the stimulated ORG signal amplitude and slope with age, meaning that cones are less reactive to the stimulus but their reaction is also slower than in the younger groups. The decrease of ORG signal amplitude with age can be explained by a few reasons. Silvestre et al [29] found that contrast-sensitivity loss in older subjects (n=20, mean = 75.9 years, SD = 4.3) is mainly due to less efficient cones absorbing four times fewer photons than young adults (n=20, mean = 26.5 years, SD = 3.79). Decrease of the pupil diameter and yellowing of the lens are other factors appearing with age that also lead to vision decline and that could also explain the decrease of the ORG signal, as these two phenomena lead to less photons reaching the retina. Other groups [16,25] have taken into account these effects and calculated the effective number of photons reaching the retina, using the power of the stimulus, the pupil diameter, the axial length and published values for the absorption of the lens and the macular pigment. They both found a linear relationship between the ORG signal and the number of photons reaching the retina. Moreover, Cooper et al [10] studied the influence of the stimulus irradiance and found that the ORG amplitude and slope increased with the stimulus irradiance. This result also supports the hypothesis that a reduction in the number of photons reaching the retina, that could be due to increased ocular media opacity in older people, is the reason why we see a reduction of the ORG signal amplitude. We chose not to correct for opacity as our aim was to get a baseline data for future patient studies, without adding any other exams. We proposed an age-dependent reference curve for the stimulated iORG signal. These results may vary depending on the specific AOSLO system used, particularly if the imaging and stimulus wavelengths differ. Our results indicate that age-matched controls may be necessary when using iORG in future clinical trials.

Funding

European Research Council 10.13039/501100000781 ( OPTORETINA(#101001841)).

Disclosures

The authors declare no conflicts of interest.

Data availability

Data underlying the results presented in this paper may be obtained from the authors upon reasonable request.

Supplemental document

See Supplement 1 ^{(2MB, pdf)} for supporting content.