Eye Tracking in Aging: Challenges, Best Practices, and Novel Frontiers

Abstract

Adults aged 65 and older are the fastest growing population demographic. Standard, stationary eye tracking has been used to test for age differences in attention and cognition toward understanding decline trajectories and identifying ways to enhance functioning into late life. More recently, advanced methods of at-home eye tracking, mobile eye tracking, and simultaneous Magnetic Resonance Imaging (MRI) eye tracking have suggested promise for insight into more dynamic and naturalistic processes of visual attentional and cognitive processing as well as their brain correlates in aging. Here, we outline challenges, best practices, and novel frontiers in the use of eye-tracking methodology among older adults. We cover considerations pertaining to optimized age-tailored study designs and procedures, eye-tracker setup as well as data quality and analysis. Throughout, we reflect on our experiences conducting novel experiments via at-home and mobile eye tracking as well as simultaneous MRI-eye tracking in aging populations. We also present empirical data comparing the quality of eye tracking in the behavioral lab vs. the MRI environment among both young and older adults, in support of simultaneous MRI-eye tracking informing brain-behavior links in aging, while acknowledging tradeoffs in data quality of this combined methodology. We propose for future research to leverage these novel, advanced eye-tracking applications for a more comprehensive and real-life capture of attentional and cognitive changes with age.

1. Introduction

Eye tracking offers a powerful technique for investigating psychological processes by approximating an individual’s gaze position in relation to an object over time. Measures derived from this approach yield insight into oculomotor behavior and most commonly include saccades and fixations. Saccades characterize eye movements [1] and fixations can be reflective of an observer’s visual attention [2, 3]. Further, pupil dilation has been suggested to capture arousal and/or cognitive effort [4]. Eye-tracking methodology has been extensively applied to the study of attention and cognition [5], socioemotional processing [6], and social interactions [7] as well as in the context of user experience in human–computer interactions [8] and as a diagnostic tool in, for example, neurodegenerative diseases (e.g., cognitive impairment [9], Parkinson’s Disease (PD) [10]).

New advances in eye tracking are now allowing researchers to address previously unattainable research questions by expanding its applications into naturalistic contexts using more “extreme” approaches. Accordingly, populations traditionally harder to recruit into lab experiments (e.g., older adults, mobility impaired or chronic pain patients, rural communities) become more accessible using at-home setups via webcams [11, 12], long-range [13], or standard stationary [14] eye trackers. Furthermore, mobile devices, which typically consist of a pair of glasses with an imbedded eye tracker and scene camera that can be worn unobtrusively in daily life, have been used to study eye gaze as individuals interact naturalistically with their environment [15, 16] or social partners [17]. Eye tracking has also started to be used in combination with brain imaging and/or brain recording techniques such as Magnetic Resonance Imaging (MRI) or electroencephalography to delineate brain-behavior links.

The continued evolution of some of these novel eye-tracking methods has also benefited their use in populations typically harder to study such as older adults, including those who experience neurodegenerative diseases. Older adults represent the fastest growing demographic segment in all industrialized nations. Individuals over the age of 65 are estimated to comprise 15.9% of the global population in 2050 and 22.4% in 2100 [18]. In addition, more and more older adults experience dramatic cognitive, socioemotional, and physical decline associated with neurodegenerative disease, with estimates of 15 million people being diagnosed with clinical Mild Cognitive Impairment (MCI) or Alzheimer’s Disease (AD) in the U.S. by 2060 [19], significantly impacting both individual lives and society at large. Eye tracking allows the capture of decline trajectories in attention and cognition, generating new insight into how to promote healthy aging.

Most of the eye-tracking research in the study of aging to date has taken place in the lab with stationary setups that facilitate standardized assessment and high data quality. These setups typically require participants to remain seated a specified distance from the testing computer and eye tracker while using a chin/forehead rest to minimize head motion. Eye tracking in this context is relatively inexpensive and straightforward to implement and has advanced understanding of visual attentional and socioemotional processes in aging [20–22] as well as has allowed for the delineation of age- and disease-related decline trajectories [23, 24]. Yet, the stationary lab setup is limited in its ability to speak to real-life phenomena. Novel applications including at-home eye-tracking setups, mobile eye tracking, and simultaneous MRI-eye tracking can capture naturalistic processes and directly link the brain and behavior.

Here, we discuss challenges and best practices for use of eye tracking among older adults (Figure 1). We cover considerations pertaining to pre-screening and participant selection as well as experimental design, data processing, and data analysis. Moving beyond the boundaries of conventional eye tracking, we then outline exciting novel frontiers via at-home setups and/or mobile devices as well as when used simultaneously with functional MRI. We report results statistically comparing spatial accuracy and data loss between eye tracking in the behavioral lab vs. in the MRI environment among both young and older adults. We finally conclude with propositions for future research that implement eye tracking at the “extremes” in research on aging.

Fig. 1.

Challenges and recommendations for use of eye tracking in older adults.

2. Challenges and Recommendations for Eye Tracking in Aging

Most modern eye-tracker systems are video-based and rely on the detection of the pupil and corneal reflection for gaze estimation. Therefore, objects such as corrective lenses or atypical morphologies that occlude the eye from the camera result in poor data quality [3]; and, critically, aging is associated with various morphological and functional changes to the eye. For instance, with age the cornea flattens, lens opacity increases, and pupil size decreases [25, 26]. Further, age-related weakening of the muscles surrounding the eyes can lead to droopy eyelids [3]. Regarding functional changes, older adults commonly experience dry eyes due to changes in the lacrimal glands and substantial decline in tear production [27–29], which can result in data loss due to increased blinking [30]. Moreover, advanced age is associated with a higher prevalence of visual impairments such as cataracts, glaucoma, and macular degeneration [31]. These eye morphological and functional changes render calibration and accurate estimation of gaze position difficult and can result in low spatial accuracy and high data loss.

A critical best practice for obtaining high quality eye-tracking data from older adults is thorough pre-screening. Important considerations include testing for visual acuity/contrast sensitivity, asking about visual impairments such as current cataracts, glaucoma, and macular degeneration, and looking for noticeable eye physiologies like droopy eyelids. This step is particularly relevant for researchers interested in using eye tracking to capture underlying attentional and cognitive processes, since any age-related changes in the eye itself would present a significant confound for the interpretation of age differences. Further, maintaining an Institutional Review Board (IRB)-approved database of “trackable” older adults from previous studies, while considering that trackability may change over time, facilitates effective participant recruitment.

Well-tailored, optimized experimental designs and research environments are essential. Task design should, for example, consider font and stimulus size and timing of stimulus presentation, which are not arbitrary decisions when considering slower visual processing and reduced visual acuity with age [32]. Many older adults wear corrective lenses, which can impair data quality by introducing additional reflection or glare; yet removal of corrective lenses may impact participants’ ability to see stimuli, resulting in squinting and subsequent periods of data loss. Use of large stimuli with high contrast and/or selection of an eye tracker that accommodates poor vision (e.g., eye trackers with interchangeable corrective lenses) will improve participants’ ability to see stimuli and preserve eye-tracker quality. We also find that using adjustable chairs is advantageous for careful positioning of participants so they can see through the lower magnification of their bifocals while minimizing glare in the eye-tracker recording. Asking participants who wear bifocals to bring their single-vision lenses to the testing session may allow for a compromise in which participants can adequately see the stimuli while reducing glare. Eye tracking is sensitive to other sources of reflection and light in the research environment, necessitating a standardized optimal environment to minimize natural light. Conducting test sessions in windowless rooms or with dark shades are solutions. Further, to reduce data loss from dry eyes and enhance participant comfort, single-use eye drops and bottled water should be provided. Allowing time for familiarization with the equipment and procedure as well as shorter study sessions with breaks can reduce fatigue and avoid discomfort from sitting upright and still for a longer time because of back or neck stiffness/pain [33]. Finally, motor control issues, such as tremors in hands and neck, can be accommodated by using forehead/chin rests and providing comfortable wrist rests as well as easy-to-use response options (e.g., joystick/keyboard/button box with large response buttons).

Other important considerations such as eye-tracker sampling rate and calibration procedure will vary as a function of study goals. The selection of sampling rates is determined by the level of precision required to answer specific research questions. For example, studies with small Areas of Interest or gaze-contingent designs benefit from higher sampling rates, whereas for studies primarily interested in fixation measures lower sampling rates may be sufficient [34]. Higher sampling rates come with tradeoffs such as more costly eye trackers and computationally expensive pre-processing due to large amounts of raw data. When selecting a calibration model, we have found that 13-point generally does not have significant advantages over 9-point calibration in stationary setups with older adults. Older eyes have the most trouble with top and bottom calibration points, especially near the corners of the screen. This is pronounced in advanced setups such as MRI-eye tracking, where spatial constraints of the scanner environment impact the light source and impair visibility of the corneal reflection and/or pupil especially around the edges of the screen. Repeating calibration/validation for challenging eyes is costly due to limited and expensive scanner time, and limiting a calibration model to the most relevant points on the screen with respect to the size and location of stimuli can result in acceptable calibration more quickly [35]. In addition to deciding on an efficient calibration model, ensuring the experimenter is well trained and experienced in troubleshooting calibration issues is a critical predictor of data quality [36], particularly with older adults.

Data should be excluded when calibration is poor or when calibration quality cannot be validated after multiple attempts, reflective of poor spatial accuracy and precision. Specific data exclusion criteria based on calibration quality, however, are typically insufficiently described in the literature and not standardized across studies, especially in novel eye-tracking applications such as simultaneous MRI-eye tracking for which data are more difficult to collect. Data loss is another important indicator of eye-tracking robustness and is defined as the proportion of raw missing samples over a recording [37, 38]. Participants with high proportions of data loss should be excluded from analysis, but as with calibration quality there is little agreement regarding in the literature what constitutes high data loss. Previous stationary eye-tracking studies in aging populations report exclusion of participants with data loss greater than 30–66% [39–41]. An important consideration involves data exclusion criteria that do not disproportionately penalize older adults, given that successful calibration is more difficult to achieve with them, as well as their relatively higher rates of data loss [41]. One solution is to oversample older adults so that individuals who need to be excluded can be replaced with others who have (more) complete/higher quality data. Further, an important issue to address is whether individuals are excluded by chance or systematically, given evidence that older adults with poor health and lower socioeconomic status have higher drop-out and exclusion rates [42, 43]. Sample selectivity can be assessed by collecting data on excluded individuals—such as demographic data like age, race, and education as well as socioeconomic status, general health, or other variables of interest—to determine whether these individuals were screened out by chance or systematically at recruitment or data exclusion stages. Although there is lack of consensus for data exclusion criteria, preregistration of exclusion criteria and accompanying rationale as well as definitions of independent and dependent variables prior to data collection improve transparency, replicability, and rigor of the approach.

Further, the selection of dependent variables must align with specific research aims [3, 34], but some eye-tracking outcomes lend themselves better than others to understanding psychological processes in aging. Table 1 summarizes common dependent variables in aging research and how these are typically calculated. For example, percent fixation duration, defined as the sum of length of fixations to an Area of Interest divided by time of fixation anywhere on the screen [44], is used frequently in the literature. This metric is advantageous for investigating visual attentional processes in relation to specified Areas of Interest and minimizes the impact of data loss over a raw fixation duration approach. Adjustments to eye-tracking outcomes such as pupil size as well as inclusion of control variables such as visual acuity/contrast sensitivity are important to account for age-group differences in eye morphology and function. For example, in the case of pupillometry, aging is associated with smaller pupil size and a restricted range of pupil dilation/constriction, requiring normalization prior to age group comparisons on cognitive load [4]. Normalization is achieved by presenting a black and white screen for 10 seconds each prior to the task to calculate minimum and maximum pupil size for each participant, which are then used to adjust pupil size for age comparisons (Table 1). Failing to normalize pupil size could result in inaccurate estimation of older adults’ cognitive load and incorrect conclusions drawn. We also recommend testing and controlling for age differences in visual acuity and contrast sensitivity such as via the Freiburg Vision Test (FrACT [45]; e.g., [46, 47]) or the Snellen test for acuity and Pelli–Robson test for contrast sensitivity [48] (e.g., [44]). These practices contribute to ensuring that observed differences in visual processing are due to genuine effects in constructs of interest and do not just represent confounds due to age-related differences in morphology and function of the eye.

Table 1.

Common Eye-Tracking Outcome Variables in Aging Research

Eye tracking variable	Functional definition	Computational definition
Fixation	Periods of time when the eye is relatively still, and an area of the visual field is focused on the fovea [2]	Sustained gaze within 1° of visual angle to Area of Interest for at least 100 ms [44] Common fixation metrics include: percent fixation duration: sum of length of fixations divided by time of fixations anywhere on-screen fixation count: sum of fixations average fixation duration: fixation duration divided by fixation count Each can be computed per trial or in aggregate across task
Saccade	Rapid eye movements that redirect fovea to new area in visual field and occur between fixations; last approximately 20–100 ms at speed of up to 900°/s with head-fixed amplitude ranging from 1.2°–90° [1]	Eye movement with velocity and acceleration threshold of 30°/s and 8,000°/s2, respectively [3] Common saccadic metrics include: frequency: sum of saccades latency: interval between target presentation and initiation of eye movement amplitude: angular distance traveled during saccade velocity: angular speed of saccade
Pupil dilation	Pupil changes size via dilation or constriction to control how much light enters eye via lens; with greater size thought to index greater arousal or cognitive load [26]	Difference between average pupil size and minimum pupil size divided by range (maximum minus minimum during normalization) multiplied by 100 [4, 47]

3. Novel Frontiers for Eye Tracking in Aging

In this section, we delve into exciting new frontiers for use of eye tracking in older adults. We will cover at-home eye tracking, mobile eye tracking, and simultaneous MRI-eye tracking approaches. We supplement our reflection on lessons learned from using these novel applications in older adults with an empirical statistical comparison of spatial error at calibration and data loss—two data quality metrics—between eye-tracking data collected in a traditional lab setup vs. in the MRI environment. The findings presented serve to inform and facilitate future research.

3.1. At-Home Eye Tracking

After many years of studying age differences in fixation to emotional stimuli using stationary eye tracking in the lab [44, 49, 50], we embarked on a project to do similar studies, but this time in their homes instead of the lab [14]. We did this for a few reasons: First, a common criticism of lab-based studies of emotional processes in aging is that the constraints of the lab also constrain the ability of older adults to display their emotional processes as they would in the familiar context of their everyday life. Second, our lab studies to date had been limited in terms of how many trials participants could complete because they were typically single-session. As part of a year-long longitudinal measurement burst study, young, middle-aged, and older adults completed in-lab eye tracking at intake and debrief and at two 5-day at-home eye-tracking periods between intake and debrief. All sessions used the same setup: a Tobii Pro X2 eye tracker attached to a laptop running iMotions software. Participants clicked through a 9-point self-calibration procedure and then completed the study. Average calibration scores from iMotions were slightly lower at home (75%) than in the lab (80%), but the amount of data retained—measured as the number of samples the eye tracker recorded per second, with a maximum of 30—was similar across both contexts (92% at home vs. 90% in lab).

We found that the in-lab data generally replicated the typical findings that older compared to young adults looked relatively less at negative Areas of Interest; however, at-home, older compared to young adults fixated relatively more on negative Areas of Interest. From these data, we concluded that the context of testing makes a difference in terms of age-related differences in fixation to emotional stimuli, though particular dimensions that matter most will be important to test directly in future research.

Most participants gave permission to be video recorded through the iMotions software during the study, so research team members could check the videos of participants who had chunks of time of missing eye-tracking data. Typical reasons for data loss were not recording failure but rather intrusion of other people in the household into the testing area, especially for conversation with the participant. This is a reminder about the logistical challenges of testing in the home environment without an experimenter present to keep participants on task.

3.2. Mobile Eye Tracking

Mobile eye tracking has key advantages beyond more typical stationary eye tracking: In particular, participants are not tethered to a particular spot and can move around in more ambulatory settings. In our experience using mobile eye tracking with older adults specifically [15], we found another advantage: Calibration was easier and more successful for a larger proportion of older adults compared to standard stationary eye tracking. Some older adults, who we could not calibrate successfully using 9-point calibration in stationary tracking, were successfully calibrated using the simple wall-based calibration for mobile eye tracking.

At the same time, we identified a key constraint to using mobile eye tracking, which is not particular to recording in older adult participants, however. Specifically, we noted that because each participant has their own unique environment as a result of moving around during eye tracking, it is not possible to use standard automated processes to filter raw gaze data into pre-determined Areas of Interest, as can be done in stationary eye tracking. The typical solution to this challenge is to do manual coding of the gazepoints superimposed on the environment video, but this approach can be extremely time-consuming depending on the nature of the stimuli. In one study, we used mobile eye tracking to investigate attentional deployment to a finite range of possible stimuli that participants freely selected on one of several computer monitors in the testing room. Given the nature of these data, we were able to transpose the 3-dimension mobile eye-tracking gaze data into 2-dimension gazepoints that then allowed us to apply standard stationary approaches to filter into Areas of Interest [15]. This difficulty with analysis and interpretation (vs. ease of data collection), however, is a major limitation of using mobile eye tracking to answer questions about age effects.

An additional consideration for mobile eye tracking is the potential for increased awareness among participants that their gaze is being monitored compared to when using standard stationary eye trackers positioned below a computer screen. Research on audience effects suggests that the perception of being observed can alter behavior [51], which may result in less spontaneous, natural gaze. To mitigate this effect, participants could be habituated to the eye-tracking glasses by wearing them for some time (“mock”) before the primary experimental task, potentially reducing the salience of the device.

3.3. Simultaneous MRI-Eye Tracking

Simultaneous use of MRI and eye tracking provides noninvasive estimation of brain activity and attentional deployment, unifying the advantages of techniques with high spatial (MRI) and temporal (eye tracking) sensitivity. Eye tracking can also complement functional MRI acquisition by allowing for verification of task compliance and whether a participant is awake during the scans. Further, simultaneous MRI-eye tracking can inform reasoning about which aspects of a stimulus during, for example, reading or face encoding give rise to neural activation [52] and allow for more direct testing of brain-behavior links. Yet, integration of these techniques is still uncommon due to challenges in hardware setup, data quality, complex data pre-processing as well as analysis, and high cost.

MRI-compatible long-range eye trackers are constructed from MRI-safe materials and work largely the same way as regular eye trackers. Setup can vary depending on the type of scanner, but typically the eye tracker and infrared light source are attached to a mount on the bore and the camera captures eye movements via a first-surface, infrared-reflective mirror (Figure 2). This mirror also displays the task to the participant by reflecting images from a projector or MRI-safe display monitor located outside of the scanner, behind the participant. The eye tracker is connected to optical cables that send data outside the shielded cabin to the experiment computer, where the experimenter conducts calibration and monitors eye movements in real-time. In many cases where multiple labs use a shared scanning facility, the eye-tracker setup is removed in between participant sessions requiring careful training and standardized procedures to ensure the setup is identical for each participant (e.g., tape markers for where the mount should be positioned).

Fig. 2.

Simultaneous MRI-eye tracking setup.

Setup complexity coupled with the extreme conditions of the MRI environment results in unique challenges to ensuring quality data. Arguably the most critical and tedious step in ensuring high-quality data is positioning the eye tracker to minimize the distance between the camera, eye, and light source from the projector (without interfering with the projection), with the goal of securing continuous recording of pupil and corneal reflection [35]. The eye is often occluded by physical properties of the scanner (e.g., size of the bore or head coils) or their shadows. In addition, it is impossible to ensure the camera is correctly positioned to record eye movements without the participant inside the scanner, which can result in participant discomfort if they must reposition themselves after being rolled into the scanner. Older adults particularly struggle with this repositioning due to reduced mobility and pain in back and neck. Further, even after eye-tracker placement and participant repositioning, the angle of the eye tracker often results in poor capture of gaze directed to lower areas of the screen since the eyelid shifts downwards slightly with the eyeball, obscuring the pupil. This issue can be exacerbated in work with older adults due to droopy eyelids or sleepiness from laying still for extended periods of time, or when participants need MRI-safe prescription goggles to see stimuli.

Together, these factors amalgamate to obscure the eye recording, often resulting in failed calibration attempts or degraded data quality over the course of the task. The time-sensitive, expensive MRI affords little flexibility in eye tracker/participant positioning as well as for recalibration. In cases where calibration cannot be validated after one or two attempts, the experimenter must skip validation, losing valuable eye-tracking data (in addition to data lost due to MRI-specific data-quality metrics such as excessive head motion, etc.), to proceed with the MRI within the strict time limit. Therefore, improvements in efficiency and precision of setup and calibration during simultaneous MRI-eye tracking are critical aspects for improving data quality.

3.3.1. Use Study: Comparison of Data Quality of Eye-Tracking Data Collected in the Behavioral Lab vs. MRI Environment in Young and Older Adults.

Unsurprisingly, previous application of synchronized fMRI and eye tracking in young adults suggests high levels of participant exclusion due to technical and calibration issues [53]. Another similar study reported removing 50% of participants due to high data loss (>45%), arguing that the extreme conditions of the MRI scanner increased blink rate [52]. In our own experience conducting simultaneous MRI-eye tracking across different age groups, we noticed greater difficulty in obtaining good calibration as well as more data loss compared to in the behavioral lab context. To quantify these differences, we directly compared data quality (i.e., error at calibration and data loss) of eye-tracking data collected in the behavioral lab with eye-tracking data collected in the MRI among the same sample of young and older adults.

Participants.

Participants in this analysis (N = 111) were a subset of participants with fMRI and eye-tracking data from a larger project (Aging OnLine Study; N = 249). The final analysis sample included 56 young (M = 21.91 years; SD = 3.69; range = 18–33 years; 80.36% female) and 55 older (M = 68.24 years; SD = 6.84 years; range = 55–81 years; 69.09% female) adults. Visual acuity¹ was worse for older adults (M = 0.27, SD = 0.09) than young adults (M = 0.21, SD = 0.01; F [1,105] = 20.24, p < 0.001); with contrast sensitivity² reduced for older adults (M = 1.58, SD = 0.30) compared to young adults (M = 1.88, SD = 0.15; F [1,108] = 45.22, p < 0.001).

We conducted sensitivity analysis using G*Power 3.1.9.7 [54] to determine the smallest effect size our study was powered to detect with the available sample size. Results from this sensitivity analysis demonstrated that with the current sample size of 111 and alpha set to 0.05, we had 80% power to detect a small effect (Cohen’s f = 0.13) [55] of the two-way interaction between Age Group and Test Environment.

All participants were recruited between April 2020 and March 2023 from the community in North Central Florida via university and internal lab participant registries, senior citizen facilities as well as churches and community centers, ResearchMatch, and word of mouth. Generally healthy women and men aged 18–100 years were eligible for participation if they were able to provide verbal and written informed consent, fluent English speakers, had a minimum 8th grade education, and were on a stable medication regimen. Exclusion criteria consisted of current major depression as indicated by a score of greater than 18 on the Beck Depression Inventory-II [56], current use of anticonvulsant, neuroleptic, sedatives, or medications known to affect cognition, uncorrected visual and hearing impairments, neurological conditions affecting the brain (e.g., major stroke, epilepsy, traumatic brain injury with loss of consciousness), unstable medical illness, significant recent cardiovascular conditions (e.g., major heart attack), severe visual impairments (e.g., glaucoma, macular degeneration, current cataract), and MRI contraindications (e.g., certain metallic objects in the body, claustrophobia).

Procedures, Equipment, and Analysis.

All study procedures were approved by the IRB at the University of Florida (IRB protocol #: IRB201801057). Only study procedures relevant to the analysis conducted here are reported (see [57, 58] for more information). The study started with thorough pre-screening to determine study eligibility and written informed consent. As part of the study, participants completed a facial trustworthiness rating task while undergoing eye tracking in the lab using a stationary eye tracker and, at a later time point, in the MRI scanner using an MRI-compatible eye tracker. Visual acuity and contrast sensitivity were assessed via the FrACT [45] during the behavioral lab session.

Procedures were identical for the in-lab and the MRI version of the task. In particular, across two task runs, participants viewed a total of 108 neutrally expressive faces that systematically varied in age (young, middle-aged, older) and sex (female, male), selected from the FACES database [59] based on perceived trustworthiness (low, mid, high) obtained from an independent sample [60]. Face presentation was pseudorandomized to ensure the same face age, sex, and perceived trustworthiness level were not presented consecutively more than two times. Each trial started with a fixation cross, presented for an average jittered duration of 2,000 ms. This was followed by the presentation of a face (image dimensions: 400 × 530 pixels) centrally over a gray background for 4,000 ms, before the rating scale appeared below the face for an additional 3,000 ms, prompting participants to rate the face from 1 = Not at all trustworthy to 10 = Very trustworthy. Information regarding eye-tracking data loss was extracted during the 4,000 ms period in which the face was presented prior to appearance of the rating scale.

Eye movements were recorded via comparable stationary and MRI-compatible SR EyeLink 1000 Plus eye trackers (SR Research Ltd., Canada) both with a temporal resolution of 1,000 Hz and a manufacturer spatial resolution of 0.02°. Participants removed bifocals or trifocals and eye makeup before undergoing eye tracking. In the lab setup, participants were positioned approximately 53 cm away from a monitor with a resolution of 1,024 × 768 and used a mounted chin rest to reduce head motion. In the MRI setup, stimuli were presented on a BOLDscreen 32 LCD (Cambridge Research Systems) with a resolution of 1,024 × 768 positioned at the end of the 70 cm bore. Participants viewed the screen through a first-surface, infrared-reflective mirror mounted to the head coil, positioned approximately 76 cm from the eye tracker. In both the lab and the MRI, each task run was preceded by a standard 9-point calibration. Calibration was accepted if “good” or “fair,” which reflected an average accuracy/spatial error less than 1.5 degrees of visual angle and a maximum accuracy/spatial error less than 2.0 degrees of visual angle. If “poor,” the calibration was repeated at 5-point and 3-point as necessary, until it was time to proceed with the rest of the MRI.

We extracted two metrics of eye-tracking data quality for analysis: (1) error at calibration and (2) data loss. Error was computed during calibration and indicates the distance between the true location of the calibration point and a participant’s estimated gaze location averaged across each calibration point (i.e., higher error was reflective of lower spatial accuracy at calibration). Data loss was calculated during the presentation of a face (before the rating scale appeared) for each participant, defined as the percentage of raw samples missing over the course of the recording [37, 38] (i.e., higher data loss was reflective of lower eye-tracking robustness) again across all trials and both task runs.

To determine differences in error and data loss between the in-lab vs. MRI assessments in both young and older adults, we conducted two separate repeated measures ANOVAs with Test Environment (in-lab vs. MRI; within-subject) and Age Group (young vs. older; between-subject) on error and data loss, respectively. The pattern of results remained the same when controlling for visual acuity and contrast sensitivity.

Data were analyzed in R, Version 4.3.1 [61], using the packages psych, Version 2.3.9 [62] and jmv, Version 2.4.11 [63], and visualized with ggplot2, Version 3.5.0 [64]. Data and analysis code are available online at the Open Science Framework: https://osf.io/9uqzv/?view_only=f468e619cc694feabafabaf74ed468cd.

Results.

Table 2 presents descriptive data of calibration thresholds (i.e., whether the calibration model was “good,” “fair,” “poor,” or unable to be validated), error, and data loss for each test environment and age group. Counts and percentages are presented for calibration thresholds, and means and standard deviations for error and data loss.

Table 2.

Calibration Thresholds, Error, and Data Loss by Test Environment for Young and Older Adults

	Lab		MRI
	Young adults (N = 56)	Older adults (N = 55)	Young adults (N = 56)	Older adults (N = 55)
Calibration thresholds (N/%)
Good	42 (75%)	39 (71%)	35 (63%)	30 (55%)
Fair	1 (2%)	2 (4%)	10 (18%)	8 (15%)
Poor	10 (18%)	7 (13%)	2 (4%)	7 (13%)
Unable to validate	3 (5%)	7 (13%)	9 (16%)	10 (18%)
Error (M/SD)	0.56 (0.30)	0.63 (0.74)	0.72 (0.27)	0.84 (0.48)
Data loss (M/SD)	6.59 (7.83)	5.72 (6.31)	17.59 (19.82)	9.89 (11.59)

Error is reported in degree of visual angle and data loss is reflected as percentage. All data were aggregated across all trials and both task runs. Note that error was extracted when calibration was validated, and data loss was only computed for the 33 young and 30 older adults with good or fair calibration during both the in-lab and the MRI.

M, mean; SD, standard deviation.

Calibration Accuracy in the Lab vs. the MRI for Young and Older Adults.

The main effect of Test Environment was significant (F [1,81] = 19.72, p < 0.001, ηp² = 0.20), with greater error during calibration in the MRI (M = 0.75, 95% CI [0.68, 0.83]) than the lab (M = 0.55, 95% CI [0.48, 0.62]) (Figure 3). Neither the main effect of Age Group (F [1,81) = 0.04, p = 0.84, ηp² = 0.0005) nor its interaction with Test Environment (F [1,81) = 2.52, p = 0.12, ηp² = 0.03) were significant.

Fig. 3.

Error during calibration in-lab vs. in the MRI for young (gray) and older (blue) adults. Error was greater for the MRI than the in-lab environment (F [1,81] = 19.72, p < 0.001, np² = 0.20). Error bars depict mean and 95% confidence intervals, each individual point represents a unique participant, and shaded areas represent distribution of data points. * represents p < 0.001.

Data Loss in the Lab vs. the MRI for Young and Older Adults.

We excluded participants with calibration that was poor or could not be validated (see Table 2) based on the literature [21, 65], resulting in 33 young and 30 older adults for this analysis. The main effect of Test Environment was significant (F [1,61] = 14.15, p < 0.001, ηp² = 0.19), with greater data loss during the MRI (M = 13.74, 95% CI [9.60, 17.89]) than in the lab (M = 6.16, 95% CI [4.35, 7.96]) (Figure 4). Neither the main effect of Age Group (F [1,61) = 2.98, p = 0.09, ηp² = 0.05) nor its interaction with Test Environment (F [1,61) = 2.87, p = 0.10, ηp² = 0.05) were significant.

Fig. 4.

Percentage of data loss in-lab vs. in the MRI for young (gray) and older (blue) adults. Data loss was greater for the MRI than the in-lab environment (F [1,61] = 14.15, p < 0.001, np² = 0.19). Error bars depict mean and 95% confidence intervals, each individual point represents a unique participant, and shaded areas represent distribution of data points. * represents p < 0.001.

Discussion.

We found that eye-tracking error at calibration was worse and data loss greater in the MRI compared to the in-lab environment for both young and older adults. These findings empirically quantify previous observations that eye tracking in the MRI is less robust than in traditional lab setups, perhaps due to difficult conditions unique to the MRI scanner (e.g., sensitivity to head motion, participant fatigue or discomfort, distraction from noise, supine position) as well as limited time for troubleshooting calibration issues. Importantly, we did not find an effect of Age Group on eye-tracking quality metrics, suggesting that our sample of older adults did not exhibit worse data quality than the young adults, providing support for feasibility of simultaneous MRI-eye tracking in older adult populations.

Our participant exclusion rates—41% young and 45% older adults—due to poor/unvalidated calibration are consistent with previous research using concurrent MRI and eye tracking in young adults, where exclusion rates were as high as 50% [52, 53]. These studies argued that such exclusion rates were justifiable, given that simultaneous MRI-eye tracking experiments are typically optimized for brain data acquisition. With this type of data collection where calibration is difficult or other technical issues impact session timing, compromises are required to ensure collection of the costly MRI data, resulting in compromises pertaining to the eye-tracking data [35]. For research on aging, procedures that prolong session duration are particularly salient as older adults often experience additional discomfort from extended time in the scanner.

Some limitations of the present study are noteworthy. First, although the study procedures were identical for the in-lab and MRI sessions, differences between the environments were unavoidable. For example, the in-lab session took place in a quiet, well-lit room, whereas the MRI environment contained the scanner noise, and the eye-tracking camera had limited exposure to the light in the room due to the position of the bore. These differences could have contributed to the poorer data quality in the MRI compared to the in-lab environment. Another limitation inherent to MRI research is sampling bias since only individuals who are safe and healthy for MRI can participate, reducing the generalizability of the findings to older adults with common health issues that make them ineligible for the MRI (e.g., certain types of metal in the body, pacemakers). Lastly, here we only reported the data quality of the eye-tracking component without considering the data quality of the MRI component (e.g., excessive head motion), which would have resulted in additional exclusions. Simultaneous MRI-eye tracking experiments are susceptible to low numbers of participants, with one study reporting only 32% of participants with complete, useable data [52]. However, despite limitations, our initial results are encouraging for researchers interested in the validity of simultaneous MRI-eye tracking among older adults. Future research efforts are needed to investigate the feasibility of this combined methodology in pathological aging, including in individuals with MCI or AD.

4. General Conclusions and Future Directions

Novel methods of at-home eye tracking, mobile eye tracking, and simultaneous MRI-eye tracking represent significant advancements in the study of aging. These applications at the “extreme” of eye tracking have demonstrated feasibility in older adults—provided that care is taken at pre-screening and age-tailored experimental design—opening exciting new avenues for future research.

An important future extension will be the inclusion of vulnerable aging populations such as individuals with MCI, AD, or PD. Neurodegenerative diseases are associated with deficits in eye movements, reflective of underlying decline in brain structure and function [66]. Abnormalities in saccadic eye movements, for example, have been found diagnostic for differentiating MCI from AD-related dementia and from healthy controls [67]. In its potential as a digital biomarker, eye tracking may be a useful for monitoring the development of cognitive disorders over time [68]. Naturalistic approaches such as at-home or mobile eye tracking can further inform understanding of how visual attentional deficits among clinical aging populations may affect activities of daily living. Mobile eye tracking, for instance, allows for assessment of navigation through the home environment, with potential for impact on the design of at-home interventions toward improving safety and independence among older adults. Further, vulnerable aging is associated with reduced activation and functional integration of key brain regions [69, 70] and simultaneous MRI-eye tracking in individuals with MCI or AD will be crucial to characterize brain-behavioral links.

Exciting future avenues also comprise mobile eye trackers in the use of dual-gaze paradigms during which eye movements of two (or more) social partners can be recorded simultaneously [71]. Advancements in machine learning for pre-processing and analysis of synchronized gaze have contributed to the growing popularity of the use of dual eye tracking in the context of face-to-face conversations [72, 73] or collaborative learning [74]. Given age-related differences in visual attention to both static [75, 76] and dynamic (e.g., videos [77], virtual avatars [78]) face stimuli, dual eye-tracking approaches promise exciting insight into cue integration (e.g., related to face emotion, eye gaze, body posture, proxemics), mutual and joint attention, as well as turn taking, and their role in social decision making in older adults.

In sum, the use and further development of novel eye-tracking techniques will significantly advance knowledge about visual attention, cognition, and socioemotional processing in aging, with potential to inform intervention to maintain and promote healthy aging.

CCS Concepts:

• General and reference → Experimentation; • Hardware → Emerging technologies; • Applied computing → Psychology;