Potential vision tester using Maxwellian view, small pupil, and different levels of wavefront correction with adaptive optics

Abstract

We demonstrate a potential vision tester (PVT) designed to study and improve the accuracy of visual acuity (VA) measurements in the aging eye. Key features include a high-resolution visual display presented in Maxwellian view, a 3 mm pupil to limit wavefront (WF) aberrations, a Hartmann Shack wavefront sensor to quantify the ocular aberrations, and a deformable mirror to correct optical aberrations. VA was measured using four alternative forced choices for a single black-on-white E stimulus in each trial, with three different levels of correction for the ocular aberrations. For normally sighted subjects, VA was not significantly better when higher-order aberrations beyond second order were corrected.

1. Introduction

Visual acuity (VA) is a fundamental measure of visual function and is often used as the primary clinical outcome measure for assessing retinal function in both clinical practice and clinical trials. However, VA measurements are influenced by the optical quality of the eye [1–4], and this dependence may introduce confounding effects on the assessment of retinal disease progression and treatment outcomes. Aging eyes undergo a variety of changes adversely affecting the optics such as decreased tear film quality [5], decreased pupil size [6], increased intraocular scatter due to increased lenticular opacification [7], and vitreous changes [8]. Some of these changes lead to increased intraocular aberrations impacting measured VAs by degrading the quality of the patient’s retinal image [8–10]. As a result, it becomes increasingly difficult to measure the visual function of the retina in a reliable and accurate manner in aging eyes without confounding optical effects.

Preserving or improving visual acuity (VA) is a primary goal of clinical trials for age-related macular degeneration (AMD) and diabetic retinopathy (DR), which are leading causes of visual impairment in the aging population worldwide [11,12]. In the USA, the economic burden of vision loss among the population older than 65 yr is 47 billion dollars [13]. Currently, there are more than 2520 clinical trials registered for AMD and 1011 trials for macular edema [14] to define the success or failure of treatment options.

Unfortunately, VA assessment methods used in clinical trials possess inherent limitations that can impact the accuracy of the results for assessing retinal health, and are more suited to documenting overall vision impairment including degradation from glare. Typically, these assessments employ traditional printed or digital charts with Newtonian viewing, which do not account for variations in pupil size that significantly affect the retinal illumination and in turn the VA. Average pupil size decreases with age for a given light level and is variable among subjects [6,15]. Additionally, optical aberrations of the eye tend to increase with age and pupil size [9,16]. As a result, both retinal illumination and image quality vary with the pupil size, contributing to variability in the VA measurements [17].

During the course of a clinical trial, participants may undergo refractive changes due to a variety of factors, including lenticular changes such as increased nuclear sclerosis, cataract surgery with intraocular lens (IOL) implants, and glycemic lens fluctuations in diabetics [18]. Additionally, there are potential posterior segment changes such as the presence of subretinal fluid in wet AMD and occasionally in diabetic macular edema that elevate the photoreceptor layer and therefore cause a hyperopic shift in the refraction [19]. These possible optical issues necessitate the evaluation of the refractive status of the eye, which is a time and resource intensive process that is often overlooked in standard VA assessments. Use of an inaccurate refraction or the habitual refraction in VA assessments can introduce significant errors in the measurements, with confidence intervals large enough to potentially affect the outcomes of clinical trials [19]. Moreover, the traditional VA methods used in clinical trials fail to measure the within subject variability, which is crucial to differentiate a true change in retinal status from measurement noise.

The use of adaptive optics (AO) techniques can potentially minimize the consequences of optical factors on measured VAs. These techniques can improve the retinal image quality by measuring and correcting the optical wavefront aberrations of the eye. One technique includes using a Hartmann-Shack wavefront sensor (HS-WF) to quantify the wavefront aberrations and a deformable mirror (DM) to compensate for these aberrations [20]. The optical wavefront aberrations represent the deviations of light from the ideal optical path, and are typically characterized by Zernike polynomials [21]. Previous works have demonstrated improvement in visual performance when optical aberrations are corrected using AO techniques [22–26]. However, these studies used pupil sizes that were larger than 3 mm to examine the impact of aberration correction on visual performance, with one goal being the enhancement of vision correction strategies. The benefits of correcting higher order aberrations on VA measurements using a 3 mm pupil remain unknown.

Because the presence of optical aberrations can significantly influence VA measurements [4], it is crucial to understand the impact of correcting aberrations. The goal is to minimize optical factors in the measurement of VA, and with a reasonable expenditure of time and money. Thus, it is desirable to avoid many of the optical issues mentioned above and to determine the importance of correction of different orders of aberrations to measured VA. To this end, we developed the Potential Vision Tester 2 (PVT2), an AO based Maxwellian view system with a 3 mm pupil. This system incorporates an HS-WF and a large stroke DM capable of correcting a wide range of lower order aberrations and higher order aberrations, eliminating the need for a typical woofer-tweeter configuration. A 3 mm pupil size is optimal as it significantly reduces optical aberrations compared to a 5 mm pupil [16] and provides the best modulation transfer function for retinal images, while avoiding the diffraction limits imposed by smaller pupils [27]. Using a visual stimulus projected through Maxwellian view with a constant 3 mm pupil size also minimizes the variability in pupil size across subjects and ensures consistent retinal illumination. In the initial description of Maxwellian view projection, Maxwell sought to increase the quantity of light reaching the retina of his eye from a point source, observing that imaging a slit on the pupil of his eye by a lens resulted in the lens being uniformly illuminated with light [28]. Imaging a light source in the eye’s pupil, instead of looking at it directly which is known as Newtonian viewing, has since been applied widely and is now known as Maxwellian viewing [29]. The strength of a Maxwellian view system is that properties of the target such as focus, size, and shape, can be controlled independently from retinal illuminance [29]. Older descriptions could not have anticipated modern compact displays, in which the point source is combined with the retinal plane target, while retaining the key design elements. In the current system there is careful separation of pupil planes and retinal planes, with the final instrument exit pupil plane imaged onto the pupil of the eye, while the retinal plane is focused by the refractive elements of the eye and ancillary optics that are placed in pupil conjugate planes. Herein, the AO components that can achieve a sharp focus of the target on the retina include the DM located in the pupil plane. The fixed pupil size is given by the optical magnification at the pupil. The AO capability allows assessment of VA measurements for accuracy, reliability, and consistency under conditions of correction of different optical aberrations, such as second order (defocus and astigmatism), second through fourth order (defocus, astigmatism, and coma), and second through sixth order.

We present the apparatus and methods implemented to study the impact of different aberration correction conditions on VA measurements and their variability. For this study we utilized a psychophysical technique that provides both the mean and the SD for each VA measurement of each subject.

2. Methods

2.1. Subjects

A total of 10 normally sighted subjects (age: 36.6 +/- 9.8 yr) participated in the VA task. All subjects had VA better than 20/25 with habitual correction and measured using a commercially available clinical VA system (SmartSystem 2, M&S Technologies, Niles, IL). This system provided a monitor-based Newtonian view VA test, demonstrating that the subjects had good VA. Optical coherence tomography (OCT) and scanning laser ophthalmoscopy (SLO) images were acquired using the Spectralis OCT II (Heidelberg Engineering Inc, Franklin, MA, USA) to provide information on retinal health. All the subjects were free from any retinal pathology as determined by an ophthalmologist. This research, conducted at the Indiana University School of Optometry was reviewed by the Institutional Review Board of Indiana University and conforms to the Guidelines of the Declaration of Helsinki on Human Subject Research.

2.2. Apparatus

To reduce artifacts due to the optics of the aging eye in VA and other visual functions, we designed a series of Maxwellian view instruments, called Potential Vision Testers (PVT) [4]. The PVT2 was designed using Optic Studio (Zemax, Ansys, Cannonsburg, PA) to obtain diffraction-limited performance for a 3 mm pupil (Figs. 1 and 2). The system used primarily reflective elements to control chromatic aberrations and unwanted reflections. Off-axis folding of concave mirrors was used [30] to reduce system aberrations. The optical design and the optomechanical components were designed using computer-aided design (CAD) modeling (SolidWorks, Dassault Systèmes, Waltham, MA, USA). The optical and mechanical designs were optimized iteratively, and the final mechanical CAD design was used as reference for optical bench setup and alignment. The system was aligned using a collimated visible diode laser (635 nm, CPS635R, Thorlabs, Newton, MJ, USA) and an ancillary portable HS wavefront sensor for calibration only (HSc) (WFS30 -7AR, Thorlabs, Newton, MJ, USA), which was positioned sequentially at specific locations to obtain precise alignment. The system included an integrated HS-WF sensor (HASO4 First, Imagine Eyes, Orsay, France) and DM (Mirao 52e, Imagine Eyes, France) as described below. These elements were optically conjugated to the pupil planes. Further, the HS-WF sensor position was optimized to ensure minimal tip and tilt in the measurements. The conjugate pupil plane of the DM at the eye was located using an ancillary USB camera sensor, and the position was used as the reference for the pupil camera alignment.

Fig. 1.

A) Schematic of the optical layout of the PVT2 system. In the visual stimulus channel (orange-shaded), a high-resolution OLED microdisplay is placed at a plane conjugate to the retina to present the stimulus, thus serving as both the source and the retinal target plane of a Maxwellian view system. Subsequent pupil and retinal planes are carefully separated. Light from the microdisplay is collimated by an achromatic lens (L1). There is a pupil plane between the two achromatic lenses (L1,L2). Next the beam passes through the second achromatic lens (L2) and a short pass dichroic mirror (SPD), forming a retinal plane prior to a fold mirror M1. Together L2 and a reflective spherical mirror (SM1) image the pupil plane of the visual stimulus channel onto the deformable mirror (DM). Subsequently, light emerging from the DM is relayed to the eye via reflective spherical mirror SM2, fold mirrors M2 and M3, spherical mirror SM3, and fold mirror M4. Focusing the retinal plane of the stimulus channel on to the retina is performed by the refractive optics of the eye and the DM, which is in a pupil plane conjugate to the pupil of the eye. The unwanted astigmatism introduced by the horizontal off axis folding of SM1 is compensated by vertical folding at SM2 and SM3. The dashed blue line indicates the HS-WF channel. An 854 nm SLD is introduced into the visual path using a beamsplitter (BS) to act as a wavefront beacon. Light returning from the eye is split from the visual channel by the SPD beamsplitter and relayed to the HS-WF sensor. To reduce corneal specular reflections, a pair of cross-polarizers (P1, P2) is placed in front of the SLD (P1) and the HS-WF (P2). Two pupil cameras, angled such that their optical axes converge at a point 80 mm along the Z-axis from their centers, monitored the eye’s pupil for alignment. B) Photograph of the PVT2 optical benchtop setup, with the OLED microdisplay at the top left and the pupil of the eye at the far right. To minimize captions, only the main components are labelled.

Fig. 2.

Spot diagrams from the Zemax model, for the stimulus channel across +/- 2 deg on the retina (3.8 mm on the display). These spots represent geometric point spread function across visible wavelengths (blue spots: 450 nm, green spots: 550 nm, and red spots: 650 nm), and the black circle represents the Airy disk.

The PVT2 configuration projected a visual stimulus in Maxwellian view through a 3 mm pupil and allowed for wavefront correction via open- or closed-loop AO measurements (Fig. 1). In open-loop operation, wavefront corrections are applied directly to the DM based on independent measurements of the eye's wavefront aberrations, without accounting for the DM's response. In contrast, closed-loop operation involves continuous monitoring of the eye’s wavefront aberrations after the DM applied corrections using the HS-WF. The wavefront correction is then iteratively optimized based on real-time feedback from the HS-WF sensor to improve correction accuracy [31]. The choice of closed-loop or open-loop operation varies according to experimental conditions and tasks, as found previously [32]. The AO subsystem in the PVT2 consisted of a HS-WF sensor to provide the WF measurements that included the human eye in combination with the system. This could be operated in either closed- or open-loop control of the large stroke DM with 52 actuators and a 15 mm effective aperture that enabled correction of defocus without an additional component such as a Badal optometer, as well as higher order aberrations.

The visual stimulus channel consisted of a high-resolution OLED microdisplay (ECX335AF, Sony Electronics, Tokyo, Japan), which filled the DM surface and was relayed with a magnification factor of 0.16x, producing a display with 4 pixels per arc min on the retina as calculated based on a Gullstrand model eye. The DM was minified 0.2 x, producing a 3 mm pupil optically conjugate to the entrance pupil of the subject’s eye. To measure the WF errors of the subject’s eye combined with the system, an 854 nm SuperLuminescent Diode (SLD) beacon was coupled to an optical fiber (Broadlighter S840, Superlum, Ireland) and collimated using a reflective collimator. The collimated beam from the SLD was combined with the visual stimulus channel before the DM and attenuated to produce an average power of 37.6 µW at the cornea.

To minimize the specular reflection from the corneal apex, a pair of cross polarizers were used. One was placed in front of the SLD and another orthogonally oriented relative to the polarization of the SLD was placed in front of the HS-WF sensor (Fig. 1). Additionally, an approximate 1 mm decentration of the beam with respect to the pupil center effectively minimized this reflection. The light reflected from the retina was split from the visual channel after the DM, using a short pass dichroic beamsplitter, then relayed with a minification from the DM, onto the HS-WF sensor producing a 3.33 mm pupil image on the lenslet array.

The HS-WF sensor and DM were controlled in closed-loop operation using manufacturer provided software (WaveTune, Imagine Eyes). The influence matrix of the DM for the control matrix was calibrated using an artificial eye consisting of an achromatic doublet (50 mm) and a diffuse reflective target. The inherent aberrations in the visual stimulus channel were measured by using the OLED display as a source to project an approximately 20 min of arc white dot on the black background, with the DM state set to flat position and recorded using the ancillary HSc. The HSc sensor was positioned at multiple collimated beam locations, such as the pupil plane in front of the system, next to the deformable mirror, and between L1 and L2 relay lenses of the display to measure wavefront quality of each optical subsystem. The major inherent final aberration of the system was residual astigmatism that resulted from the off axis folding design coupled with the need to meet the mechanical constraints for practical human testing. The HS-WF channel shared a common path with the visual stimulus channel except at the final relay lens. Therefore, wavefront aberrations measured by HS-WF in the system accounted for most of the inherent system aberrations along with the ocular measurements. The DM correction in the closed-loop operation effectively corrected the inherent system aberrations along with the ocular aberrations. The residual aberration in the non-common path portion of the system between the display and SPD arose from the use of on-axis achromats (L1, L2) and produced a root mean squared (RMS) wavefront error of 0.017 μm.

2.3. Wavefront correction

Each subject performed the VA task under three different WF aberration correction conditions, in order of test: 1) correction of Zernike second order aberrations (Z₂), which included Z_{2, -2,} Z_2,0, Z_{2, 2}; 2) correction of Zernike second through fourth order aberrations (Z₂-Z₄); and 3) correction of Zernike second through sixth order aberrations (Z₂-Z₆). The three conditions were repeated to assess the repeatability of VA measurements under these conditions. The experiment was run with subjects who had little previous experience, and the WF correction with feedback took a variable amount of time. Thus, to ensure that we could complete all six datasets before the subject was tired, we omitted our usual practice trials. The lack of practice for both the correction procedure and the VA test was anticipated to have the greatest impact on the first measurement, i.e. correction of the first Z₂ data. The subject’s head was stabilized using a head and chinrest. Precise pupil alignment was achieved using the 3D motorized headrest, and real-time feedback from the pupil cameras. During the pupillary alignment and wavefront beacon centration, subjects fixated on an approximately 20 min of arc fixation cross with the DM state set to the flat position.

Ocular aberrations were measured using the HS-WF at a rate of 15–30 Hz across a 3 mm pupil. Baseline ocular wavefront measurements were recorded at the start of each trial with the DM set to a flat state. A closed-loop correction was applied using the DM for each correction condition until the residual RMS error reached a stable minimum. Subjective feedback was acquired, as the residual RMS error stabilized. Once RMS error was at a stable minimum the closed-loop correction was paused, and the DM state was held fixed. The residual aberrations were recorded, and the wavefront beacon was turned off by closing a shutter (VS14, Uniblitz, Rochester, NY, USA). When present the beacon was visible as a dim red light. All the wavefront measurements (RMS errors), and DM corrections excluded tip, tilt and piston aberrations. The VA task was performed with the DM state set to the correction. The pupil position was monitored as the subject performed the task. The wavefront was monitored again at the end of each VA measurement to ensure proper correction was achieved.

2.4. Visual acuity task

VA measurements were acquired using a one up one down adaptive staircase method. The VA stimulus consisted of a black letter “E” on a 2.5 deg white background, presented in a four alternative forced choice paradigm (up, down, left, or right). Subjects were allowed continuous viewing of the E until a response was recorded. A 20/40 letter was presented initially, and letter size changed in small steps: a 1 minimum angle of resolution (MAR) step for sizes >20/30 and 0.5 MAR step for sizes <20/30. The letter sizes ranged from 20/10 to 20/400, with a total of 40 presentations per condition to model how the number of trials impacts measurement performance. The test was terminated if the subject provided three consecutive correct responses at the 20/10 level, which corresponded to an E with a 2 pixel spacing between features. In cases where subjects responded correctly to all stimuli, VA was considered as 20/10.

The VA stimulus was presented using a custom MATLAB program based on Psychtoolbox-3 [33]. The responses to each E were fit with a cumulative Gaussian function using a minimum root mean squared error criterion, which provided the mean (50% threshold) and standard deviation (SD) of the VA measurement for each individual subject, using a custom MATLAB program (Mathworks, Natick, MA, USA) [4,19].

Statistical analyses such as ANOVA and paired t-tests were performed using SAS studio (SAS Institute, Cary, NC, USA). Four separate two-way repeated-measures ANOVAs with factors time 1 vs. time 2 (2 levels) and WF correction condition (3 levels) with Bonferroni correction were utilized. We computed the coefficient of variation (CV) as SD/Mean (50% threshold VA) from the cumulative Gaussian fits. The residual RMS error is the residual wavefront error after the wavefront correction was applied. The four ANOVAs were performed on the 50% threshold VA, SD, CV, and residual RMS error. We further explored the differences between the correction condition for time 1 and time 2 in separate one-way ANOVAs, but none of the analyses reached statistical significance. The significance level of 0.05 for the p-values was used for all statistical tests.

3. Results

3.1. Overall mean, and SD

The refractive errors of the subjects ranged from a spherical equivalent of –4.25 to +1.25 D (Table 1). The total baseline RMS wavefront error of the subjects, as obtained in preparation for the first Z2 correction condition, ranged from 0.294 to 1.312 microns, as in Table 1. Eight subjects had minimal Z2 aberrations, i.e. spherical and astigmatism (spherical equivalent: -0.62 to 1.25 D), while two had Z2 aberrations that were larger (spherical equivalent: -4.25 D, -1.92 D).

Table 1. Subject information.

Subject No	Gender	Age (yr)	Refractive Error (Sphere / Cylinder * Axis)	Baseline RMS Error (Microns)
F1074	Female	28	0.00 / -0.25 * 007	0.416
M851	Male	29	0.75 / -0.50 * 006	0.592
F1072	Female	30	0.50 / -1.00 * 001	0.382
F1037	Female	32	1.00 / -0.50 * 129	0.294
F1073	Female	33	0.25 / -0.25 * 012	0.483
M850	Male	33	-0.75 / -2.35 * 098	0.520
M849	Male	34	-0.25 / -0.75 * 119	0.332
F804	Female	39	0.25 / -0.25 * 158	0.391
F007	Female	46	-3.5 / -1.5 * 090	1.31
F636	Female	62	1.75 / -1.25 * 040	0.332

The VA data were well-fit by the cumulative Gaussian function. The variability of responses was similar across the correction conditions (Fig. 3). The VAs of individual subjects were excellent across the correction conditions, as measured by the 50% threshold VA expressed in MAR. The Mean +/- SD of the overall data from both time 1 and time 2 measurements were as follows: Z₂: 0.76 +/- 0.28, Z₂-Z₄: 0.77 +/- 0.23, Z₂-Z₆: 0.79 +/- 0.13 (Figs. 4 and 5). The 50% threshold VA was not significantly different across the correction conditions (Two-way ANOVA, df = 2, F = 0.11, p = .298 with Bonferroni correction). Figure 6 shows the 50% threshold VA as function of residual RMS error for the 10 subjects, plotted for all three correction conditions including time 1 and time 2. The positive correlation between VA and residual RMS error was not statistically significant (VA = 1.21 × Residual RMS + 0.644, r² = 0.0196, p = .881).

Fig. 3.

The VA data of a 32-yr old female subject (F1037) for the Z₂ (left), Z₂-Z₄ (center), and Z₂-Z₆ (right) correction conditions. The orange dots represent the proportion of correct responses across the presented letter sizes (MAR). Z₂: 50% threshold VA (0.93 MAR), SD (0.22 MAR), Z₂-Z₄: 50% threshold VA (0.72 MAR), SD (0.43 MAR), and Z₂-Z₆: 50% threshold VA (0.76 MAR), SD (0.25 MAR). The blue line represents the cumulative Gaussian fit to the VA data.

Fig. 4.

Association between VA measurements and residual RMS error for the first trial across different wavefront correction conditions. The top row shows the 50% threshold VA (MAR) as a function of residual RMS error (microns) for Z₂ (left column), Z₂-Z₄ (center column), and Z₂-Z₆ (right column) correction conditions. The center row shows the SD (MAR) as a function of residual RMS error (microns), and the bottom row shows the CV as a function of residual RMS error (microns) across all the wavefront correction conditions. Each colored dot represents an individual subject, and the dotted line represents the trend line (statistically not significant). For Z₂-Z₄ correction condition (top center), the 50% threshold VA had a significant positive correlation with residual RMS error shown as solid trend line (VA = 5.01 x Residual RMS + 0.243, r² = 0.729, p = 0.019).

Fig. 5.

Association between VA measurements and residual RMS error for the time 2 measurements across different wavefront correction conditions. The top row shows the 50% threshold VA (MAR) as a function of residual RMS error (microns) for Z₂ (left column), Z₂-Z₄ (center column), and Z₂-Z₆ (right column) correction conditions. The center row shows the SD (MAR) as a function of residual RMS error (microns) and the bottom row shows the CV as a function of residual RMS error (microns) across all the wavefront correction conditions. Individual subject data are denoted by colored dots, with each subject's color code consistent with the representation used in Fig. 4 for ease of comparison. The dotted line represents the trend lines (statistically not significant).

Fig. 6.

The 50% threshold visual acuity (VA) plotted as a function of residual RMS error in microns across different correction conditions. The black, blue, and red dots correspond to VA for the Z_2, Z₂- Z₄, and Z₂-Z₆ correction conditions, respectively.

The VA data were well-fit by the cumulative Gaussian function. The variability of responses was similar across the correction conditions (Fig. 3). The VAs of individual subjects were excellent across the correction conditions, as measured by the 50% threshold VA expressed in MAR. The Mean +/- SD of the overall data from both time 1 and time 2 measurements were as follows: Z₂: 0.76 +/- 0.28, Z₂-Z₄: 0.77 +/- 0.23, Z₂-Z₆: 0.79 +/- 0.13 (Figs. 4 and 5). The 50% threshold VA was not significantly different across the correction conditions (Two-way ANOVA, df = 2, F = 0.11, p = .298 with Bonferroni correction). Figure 6 shows the 50% threshold VA as function of residual RMS error for the 10 subjects, plotted for all three correction conditions including time 1 and time 2. The positive correlation between VA and residual RMS error was not statistically significant (VA = 1.21 × Residual RMS + 0.644, r² = 0.0196, p = .881).

The SDs of the VA from the individual subject’s cumulative Gaussian fits were not significantly different across the correction conditions (Two-way ANOVA, df = 2, F = 0.22, p = .267 with Bonferroni correction) (Figs. 4 and 5). The CVs were not significantly different across the different correction conditions (Two-way ANOVA, df = 2, F = 0.22, p = .259 with Bonferroni correction).

3.2. Variability of VA among repeated measures

A two-way ANOVA revealed no significant differences in 50% threshold VA measurement among repeated measurements, across the correction conditions (5 df, F = 0.33, p = .894). A paired t-test further indicated no significant difference between the repeated measurements across all correction conditions (Z₂: p = .279, Z₂-Z₄: p = .791, and Z₂-Z₆: p = .770). This result suggests that VA measurement showed reliability across different correction conditions. The residual RMS error remained consistent between repeated trials across all the correction conditions (paired t-test: Z₂: p = .383, Z₂-Z₄: p = .842, and Z₂-Z₆: p = .268).

However, despite the consistent residual RMS error between repeated measurements, the SDs of VA were significantly larger for the time 1 measurements with the Z₂ correction condition compared with the time 2 measurements (paired t-test, p = .0379). Data were collected with the Z₂ correction for the time 1 measurements without prior practice trials, whereas the time 2 measurements followed two additional wavefront correction conditions and 80 trials of VA measurements. No significant differences were found for the Z₂-Z₄ (paired t-test, Z₂-Z₄: p = .108) and Z₂-Z₆ (paired t-test, p = .767) correction conditions. The CV was significantly larger for the time 2 measurements for the Z₂-Z₄ correction condition compared to time 1 measurements (paired t-test, p = .0452). No significant differences were observed for the other correction conditions (paired t-test: Z₂: p = .0562, Z₂-Z₄: p = .0425, and Z₂-Z₆ : p = .599).

ANOVA comparisons performed on each of the VA measurements (50% threshold, SD, and CV) separately for time 1 and time 2 data were not significantly different across the correction conditions either in time 1 or time 2 measurements. A significant positive correlation was observed between the VA and the residual RMS error only for time 1 measurement with the Z₂-Z₄ correction condition (VA = 5.01 x Residual RMS + 0.243, r² = 0.729, p = 0.019). However, none of the other VA measurements had any significant correlation with the residual RMS error (Figs. 4 and 5, Table 2). There were substantial individual differences in SD and CV, with a subject M850 having unusually large SD’s both time 1 and time 2 (Fig. 7). In contrast, subject F007 also had larger wavefront errors than typical for this group of subjects, but not atypical 50% threshold, SD, and CV.

Table 2. Linear regression fit data of the VA measurements (y = a * Residual RMS + b) as a function of residual RMS error across the correction conditions.

Condition	VA (y)	Trial	Mean (MAR)	Mean Log(MAR)	SD (MAR)	Slope (a)	b	R^2	P
Z2	50% Threshold (MAR)	1	0.70	-0.46	0.34	4.1	0.22	0.063	0.86
		2	0.82	-0.22	0.21	0.47	0.77	0.0040	0.91
	SD (MAR)	1	0.41	-	0.11	3.7	-0.026	0.12	0.74
		2	0.19	-	0.22	-2.1	0.46	0.33	0.35
	CV	1	0.87	-	0.93	-2.1	1.1	0.0021	0.99
		2	0.23	-	0.11	-3.0	0.60	0.60	0.067
Z2-Z4	50% Threshold (MAR)	1	0.78	-0.27	0.16	5.01	0.24	0.73	0.016
		2	0.76	-0.33	0.30	-0.70	0.83	0.0021	0.99
	SD (MAR)	1	0.29	-	0.14	0.55	0.23	0.011	0.97
		2	0.39	-	0.14	-0.77	0.47	0.011	0.97
	CV	1	0.39	-	0.21	-1.7	0.57	0.045	0.90
		2	0.55	-	0.21	-0.065	0.55	0.00038	0.99
Z2-Z6	50% Threshold (MAR)	1	0.79	-0.25	0.10	0.15	0.77	0.0019	0.99
		2	0.80	-0.24	0.15	0.40	0.76	0.0031	0.99
	SD (MAR)	1	0.31	-	0.13	-3.3	0.64	0.48	0.16
		2	0.34	-	0.33	-6.8	0.98	0.19	0.60
	CV	1	0.40	-	0.16	-4.0	0.80	0.52	0.12
		2	0.52	-	0.70	-13	1.8	0.16	0.66

Fig. 7.

Higher order Zernike coefficients of the ocular wavefront error of three subjects (M850, F007, F1037). M850 is the 33-yr old male subject with the most inconsistent performance across the correction conditions. The 50% threshold +/- SD of VA for Time 1 were Z₂ : 0.68 +/- 0.14; Z₂-Z₄: 1.1 +/- 0.34; Z₂-Z₆: 0.79 +/- 0.29, for Time 2 were Z₂ : 1.3 +/- 0.39; Z₂-Z₄: 1.5 +/- 0.69; Z₂-Z₆: 1.1 +/- 0.41. F007 is a 46-yr old female subject with largest RMS wavefront error among the 10 ten subjects, but with better performance on VA than subject M850. Her 50% threshold +/- SD of VA for Time 1 were Z₂ : 1.3 +/- 0.22; Z₂-Z₄: 0.76 +/- 0.41; Z₂-Z₆: 0.70 +/- 0.31, for Time 2 were Z₂ : 0.91 +/- 0.21; Z₂-Z₄: 0.57 +/- 0.48; Z₂-Z₆: 0.73 +/- 0.31. In comparison a typical subject F1037 is the 32-yr old female subject shown in Fig. 3. The 50% thresholds +/- SD of VA for Time 1 were Z₂ : 0.52 +/- 0.33; Z₂-Z₄: 0.79 +/- 0.29; Z₂-Z₆: 0.66 +/- 0.36, for Time 2 were Z₂ : 0.94 +/- 0.23; Z₂-Z₄: 0.72 +/- 0.43; Z₂-Z₆: 0.77 +/- 0.26.

4. Discussion

Accurate measurement of VA is essential for evaluating retinal function and assessing the efficacy of treatments for retinal diseases. However, traditional VA measurement methods, particularly those used in clinical trials, are often influenced by a patient’s optical conditions that change over time as well as suboptimal refractive correction, which can confound the assessment of visual function. In this study, we introduced the PVT2, an AO-based Maxwellian view system with a 3 mm pupil, designed to evaluate the impact of different levels of wavefront aberration correction on the accuracy and reliability of VA measurements. By mitigating optical artifacts, the PVT2 provides a clear distinction between optical and retinal contributions to VA, in order to improve measurement accuracy and repeatability.

Our results demonstrated that in 10 normally sighted subjects, VA remained similar across the three aberration correction conditions: Z₂, Z₂-Z₄, and Z₂-Z₆, establishing that the 3 mm pupil size is sufficiently small to produce adequate image quality with correction of only defocus and astigmatism. Furthermore, the VA measurements demonstrated good repeatability and didn’t vary between repeated measurements across the correction conditions. For the time 2 measurements, the Z₂ aberration correction condition has the lowest CV. The higher CV for time 1 compared to time 2 in the Z₂ aberration correction may have resulted from this being the first dataset collected and the lack of practice trials, including both obtaining the correct subject spherical adjustments and the VA test procedure. Also, the poor performance of an individual subject may have contributed to more variability at time 1 and the positive correlation between the residual RMS and the SD for the time 1 measurements alone. More subjects and a wider range of aberrations could provide a clearer understanding of factors leading to poor performance. Overall, our results suggest that a Maxwellian view system with a 3 mm pupil provided accurate and reliable VA measurements, since correction of higher-order aberrations (Z₂-Z₄ and Z₂-Z₆) did not result in a significant improvement in VA compared to correcting only second-order aberrations (defocus and astigmatism, Z₂). These results align with the fact that higher order ocular aberrations are minimized for a 3 mm pupil [10,16,27].

The FDA recommends use of VA as the primary outcome measure for evaluation of the treatment efficacy of therapeutics for retinal diseases [34]. A commonly used criterion for VA change is an improvement or decline of more than 15 letters (>0.3 Log(MAR)) [35]. However, a threshold of more than 5 letters (>0.1 Log(MAR)) is also used [36,37]. Additionally, a non-inferiority margin of 3 - 5 letters (0.06–0.1 Log (MAR)) or less is often applied to determine whether two treatments produce comparable outcomes [38]. The traditional VA measurements are known to have poor reproducibility, often varying more than 0.1 Log(MAR) [39]. This variability can obscure the true change in VA due to an improvement or worsening of the retinal status. In AMD patients, exudation commonly occurs, and photoreceptor elevation occurs in the presence of subretinal fluid. If habitual refraction is then used, there is a resulting hyperopic defocus. Our recent study demonstrated that the resulting hyperopic shift induced by photoreceptor elevation can result in VA confidence intervals larger than 0.1 Log(MAR) [19], potentially introducing error in the assessment of treatment effects and complicating the interpretation of efficacy.

Recent clinical trial results of a newly approved photobiomodulation therapy for AMD raised concerns, as the treatment group demonstrated improvement in best corrected VA of only approximately 0.11 Log(MAR), compared to 0.06 Log(MAR) change in the sham treatment group [36,40]. Moreover, clinical trials often fail to provide individual subject variability in the VA measurements, making it difficult to assess the measurement noise in the study. Such limitations can severely complicate clinical decision making, as the true treatment effect remains uncertain, and increase the cost of clinical trials because very large sample sizes are required to overcome the issues of high variability.

A precise and reliable VA measurement tool with reduced measurement noise, i.e. low SDs, is essential for accurately determining true treatment effects. The optical design of PVT2 offers several advantages in reaching this goal. Firstly, incorporating a 3 mm pupil effectively minimized the optical artifacts due to variability of pupil size, and the effects of the increased ocular aberrations with larger pupils on the VA measurement. Secondly, implementation of the AO with a large stroke DM provided precise correction of both sphere and cylinder as well as higher order optical aberrations, resulting in retinal image quality largely free from these optical effects of the eye. Our results showed that in this healthy population there was no benefit in correcting a subject’s aberrations beyond Z₂. Lastly, the adaptive staircase approach improved the accuracy of VA measurements by providing psychometric data and quantifying the measurement variability of VA at the individual subject level. This VA testing method [4,19,24,25] is widely known among visual scientists, but is not the basis of the devices most often used in clinical trials [41,42 ]. The VA task requires only about 2-3 min. Only the Z₂ correction condition is needed to obtain acceptable results in the subjects tested thus far. Thus, there could potentially be an improvement in accuracy and also a removal of the bottleneck made worse by not permitting pupil dilation or autorefraction corrections for testing VA in clinical trials [41,42] and management of patients with retinal disease.

While this study provides valuable insight on reliability of VA measurements under different aberration conditions, the main limitation is the relatively small sample size of 10 subjects. All the subjects participating in this study were normally sighted and may not represent the effects of aberration correction in an older clinical population with AMD or the DR population with potentially increased lens alterations. We plan to study these effects in aging subjects over a wider range of aberrations conditions, with presence of lenticular changes, IOL implants, diabetes, and retinal pathology. It is notable that those optical issues of the aging eye which increase scattering and therefore reduce contrast at the retina, such as tear film breakup, and lenticular and vitreous opacities in the optical path of the eye, are not corrected by AO and still could contribute to an optical impairment of the subject’s VA. Overall, these results demonstrate that the use of a Maxwellian view system with a 3 mm pupil could potentially improve the VA measurement accuracy in a reliable manner, even with just correction of lower order aberrations.

Funding

National Eye Institute 10.13039/100000053 ( EY030829).

Disclosures

VP: none, AEE: Aeon Imaging, LLC (I,P), SAB: none, TJG: Aeon Imaging, LLC (E).

Data availability

Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.