Ratiometric analysis of in vivo optical coherence tomography retinal layer thicknesses for detection of changes in Alzheimer's disease

We analyzed ophthalmic retinal optical coherence tomography (OCT) images from patients with Alzheimer's disease (AD) to identify retinal layer thickness and ratio changes that may serve as image‐based biomarkers for the disease. One three‐dimensional volume before and one after diagnosis for each of 48 patients were segmented to identify retinal layer and total retinal thicknesses. Between before‐ and after‐diagnosis retinal OCT images, there were significant thickness changes in six of 10 (60%) retinal layers across all 48 patients. Through a comparison with age‐matched healthy subjects, the significant changes were attributed to AD only (NFL and PR2 layers), age only (GCL, IPL, and RPE layers), or both AD and age (OPL layer). Analyzing ratios of retinal layer thicknesses, 53 of 90 (58.89%) ratios had significant changes. The four independently nonsignificant layers were assessed to be affected by neither AD nor age (INL layer) or both AD and age (ELM, PR1, and BM layers). The demonstrated image segmentation, measurement, and ratiometric analysis of retinal layers in AD patients may yield a noninvasive OCT image‐based retinal biomarker that can be used to detect retinal changes associated with this disease.


K E Y W O R D S
Alzheimer disease, biomarkers, early diagnosis, optical coherence, retina, tomography

| INTRODUCTION
Alzheimer's disease (AD), the most common neurodegenerative disorder, is a significant medical, societal, and financial burden for a large proportion of the population worldwide and remains without a cure [1].AD is the leading cause of dementia, a group of conditions characterized by impaired memory, thinking, communication, and behavior [2].In 2015, an estimated 47 million people worldwide were living with dementia, with AD accounting for 60% to 70% of cases [3,4].Symptoms of AD begin with slow-onset memory impairment and progressive disorientation, language disturbance, loss of visuospatial skills, and changes in behavior and personality [1].Over the course of the disease, as cognitive decline worsens, neuropsychiatric symptoms can be observed [1].A diagnosis of AD requires the presentation of specific pathological brain changes, such as the appearance of beta-amyloid (Aβ) plaques and neurofibrillary tau tangles, preceding the onset of symptoms classically seen in dementia [2].
Current biomarkers for the pathogenesis of AD include those of Aβ deposition, neuronal injury, and associated biochemical changes such as oxidative stress [5].Analysis of biomarkers contained in the cerebrospinal fluid (CSF) such as CSF Aβ 42 requires a lumbar puncture, an invasive and painful surgical procedure [6].Other biomarkers can be analyzed with less invasive methods, such as imaging techniques including positron-emission tomography (PET) and single-photon emission computerized tomography (SPECT) perfusion imaging [5].Technological advances in hyperspectral imaging and adaptive optics have allowed imaging modalities such as scanning laser ophthalmoscopy (SLO) to identify retinal gliosis indicative of AD pathology [7].Optical coherence tomography (OCT) is a noninvasive optical biomedical imaging technique that can provide high-resolution label-free cross-sectional images of retinal microstructure and its neuronal layers (Figure 1) [8][9][10][11][12].A recent seminal study analyzed ophthalmic OCT images from AD patients and suggested there are decreased thicknesses of the nerve fiber, ganglion cell, and inner plexiform layers in the retinas of AD patients [13].Retinal nerve fiber layer thicknesses were measured in another study using OCT in AD patients, and although the authors found nonsignificant thinning compared to healthy controls, they suggested this could be due to a small AD cohort size and limitations of the OCT system to assess retinal layers [14].A systematic review of 71 studies such as this found that while AD patients often exhibit retinal thinning on OCT retinal scans, the literature lacks sufficient evidence on the specific areas of the retina that undergo changes [15].Another review of 64 studies focusing on biomarker identification using OCT in patients with AD concluded that a very limited number of studies analyze patients with preclinical AD or are longitudinal, arguing that conducting studies following the evolution of preclinical AD patients would be of great value [16].These identified gaps in the literature motivate a longitudinal study of retinal layer changes in AD patients.
There is utility to expand this field of research as the discovery of noninvasive OCT image-based retinal biomarkers would have significant clinical and societal F I G U R E 1 Representative retinal layer-segmented optical coherence tomography images acquired from the macular region.Images are captured from a patient before and after their diagnosis of Alzheimer's disease.BM, Bruch's membrane; ELM, external limiting membrane; GCL, ganglion cell layer; ILM, internal limiting membrane; INL, inner nuclear layer; IPL, inner plexiform layer; NFL, nerve fiber layer; OCT, optical coherence tomography; OPL, outer plexiform layer; PR1, layer of the inner segment or ellipsoid zone; PR2, layer of the outer segment or interdigitation zone; RPE, retinal pigment epithelium.
impact.A large clinical study showed that a thinner retinal nerve fiber layer was associated with an increased risk of developing dementia, including AD, and advocated the use of OCT as a noninvasive screening tool [17].OCT has become a standard for the detection and diagnosis of many retinal diseases (eg, macular edema and holes), as well as retinal manifestations of systemic diseases (eg, diabetic mellitus [DM]) [18].
OCT is used extensively to assess variations in retinal thickness, such as thinning due to age-related macular degeneration (AMD) and thickening due to diabetic macular edema [19].Ratiometric analyses have been recently introduced to investigate subtle changes in OCT-measured retinal layer thicknesses, and the identified ratios have been considered more sensitive parameters for early detection of diseases including diabetic retinopathy and multiple sclerosis [20][21][22].In this study, we aim to identify in vivo retinal thickness changes in individual layers and in ratios of layer thicknesses that may be associated with Alzheimer's disease.Through novel ratiometric analysis of retinal layers in ophthalmic OCT images, we demonstrate the potential to noninvasively identify patterns of retinal layer changes as a biomarker of AD.

| METHODS
The human subjects research protocol for this study was reviewed and approved by the Institutional Review Boards at Carle Foundation Hospital and the University of Illinois at Urbana-Champaign.The research adhered to the tenets of the Declaration of Helsinki and HIPAA standards.All data was gathered, stored, and kept confidential in a secure electronic format.

| Human subject data
Subject recruitment involved data mining under waivedconsent, where we queried the electronic medical record database for patients matching our search criteria.Our search identified patients seen at Carle Foundation Hospital Ophthalmology Clinic between 1 January 2012 and 1 January 2020, who had been diagnosed with Alzheimer's disease and had at least one OCT imaging session both before and after diagnosis.To control for confounding retinal layer thickness changes, additional search criteria were applied to identify and exclude patients with histories of retinal disease, neurological disease, or stroke, but none were identified.Search criteria to exclude patients with family histories of AD were applied, but due to the data being retrospectively mined, providers did not have specific instructions to collect AD family history, and thus it was not available as data for use in filtering subjects.With the applied inclusion and exclusion criteria, we ensured the 121 AD subjects identified had their OCT images collected during routine ophthalmic checkups at the clinic, during visits unrelated to their AD diagnosis.As the image capture settings varied between subjects and sometimes across images acquired from a single subject during a session, we manually filtered the 121 subjects to include only those subjects who had OCT scans within the macula, and at least one pair of imaging sessions for the same eye, in order to facilitate comparisons between similar regions at the two different time points.The filtering resulted in 48 subjects identified as having these criteria, providing a total of 96 OCT imaging sessions.A flowchart of the study design is depicted in Figure 2. The demographic and age data of the 48 subjects in the study are listed in Table 1.

| Retinal imaging and segmentation
All retinal imaging was performed using a commercial ophthalmic OCT system (SPECTRALIS Optical Coherence Tomography, Heidelberg Engineering Inc.), which F I G U R E 2 Flowchart of subject and image selection.Application of inclusion and exclusion criteria for subject selection followed by filtering subject images resulted in the number of segmented images listed.AD, Alzheimer's disease; OCT, optical coherence tomography offered an optical resolution of 7 μm axially by 14 μm laterally.All data was stored locally in the clinic on the computer associated with the system.
From each OCT imaging session, a three-dimensional volume of data was acquired centered on the macula of the retina, containing the fovea.Each 3D volume consisted of 241 B-scans and each B-scan consisted of 1153 A-scans.The B-scans represented two-dimensional crosssectional images of the macula arranged laterally together to produce the volume, while the A-scans represented adjacent columns that comprised a cross-sectional B-scan.
A commercial retinal layer segmentation program ("Layer Segmentation Export", Heidelberg Engineering) was used to segment and measure each retinal layer in each B-scan of the volume.Representative segmented images in Figure 1 display a macular region of the retina, distant from the fovea.All segmentation measurements were given as a vector of distances from the top border of the captured image at each A-scan position to the bottom of the retinal layer in question.The anatomical region between the top border of the captured image and the internal limiting membrane was the vitreous, and this distance was subtracted off before taking measurements of the retinal layer thicknesses.Thus, each retinal layer thickness was calculated by subtracting off the thickness of the layer immediately above it in the image.This subtraction occurred at every A-scan within each B-scan, for the entire three-dimensional OCT volume.The mean ± SD retinal layer thickness for each of the layers in the volume captured during an imaging session was therefore determined by calculating the mean ± SD layer thickness for all the A-scans in each B-scan, then averaging across all of the B-scans in the volume.This mean ± SD also represents a global average of the retina centered on the macula, which also includes the fovea.
For some B-scans, several of its constituent A-scans would contain a mixture of numerical data and "NA" empty values.This was due to the inability of the software to determine numerical values if the signal-to-noise level of the image was low, and the intensity difference between layers was not sufficient.Due to the unreliable nature of A-scan segmentation measurements in a B-scan image containing any number of "NA" empty values, the entire B-scan containing "NA" empty values was removed from consideration in determining the average global retinal layer thicknesses.

| Retinal layer analysis
Retinal thicknesses are useful to measure and directly compare as they often are time-dependent indicators of changes in pathophysiological processes.Ten layers of the retina were measured for this analysis, including the nerve fiber layer (NFL), ganglion cell layer (GCL), inner plexiform layer (IPL), inner nuclear layer (INL), outer plexiform layer (OPL), external limiting membrane (ELM), layer of the inner segment or ellipsoid zone (PR1), layer of the outer segment or interdigitation zone (PR2), retinal pigment epithelium (RPE), and Bruch's membrane (BM).Additionally, the total retinal thickness from the top of the NFL to the bottom of the BM was calculated by summing the thicknesses of the 10 individual layers.
Comparisons of retinal thickness ratios have additional utility as they can be more sensitive indicators of changes, as indicators of correlations associated with changes between two layers, and potentially as internal controls that reduce the inter-subject variability.We have previously demonstrated the use of ratiometric retinal layer analysis for changes associated with diabetic retinopathy and multiple sclerosis [20,21].In our ratiometric analysis for this study, we calculated 90 ratios between the thicknesses of individual retinal layers.To do so, we generated 45 A/B ratios of retinal layer thicknesses where A and B are two distinct layers.Each of the A/B ratios has an inverse B/A ratio that involves the same two layers but is mathematically unique.We compared these 90 retinal layer thickness ratios from the before-diagnosis data to those from the after-diagnosis data.
We have also included and analyzed OCT-measured retinal layer thickness data from a study involving healthy subjects of similar age to those included in this study [19].These results were used to determine whether changes in retinal layer thicknesses were primarily ADrelated, age-related, both AD-and age-related, or neither AD-nor age-related.In cases where no significant changes were observed in age-matched healthy subjects but were observed in AD subjects, changes were attributed to AD.In cases where significant changes were observed in age-matched healthy subjects but were not observed in AD subjects, changes were attributed to age.
In cases where no significant changes were observed in either age-matched healthy subjects or aging AD subjects, individual retinal layer thicknesses were analyzed.If changes could still not be clearly attributed to either AD or aging, comparisons of retinal thickness ratios were used to determine the likely contributor.

| Statistical analysis
The statistical analyses comparing retinal layer thicknesses and retinal layer thickness ratios were performed using the statistical computing software R (R Foundation for Statistical Computing, Vienna, Austria).Paired t-tests were conducted between the before-diagnosis and afterdiagnosis OCT data for each subject.Unpaired t-tests were conducted between the cohorts from the literature, Age-related differences (μm) were calculated from retinal layer thicknesses reported in the literature to determine the effect of age-related changes alone, and compared to the differences measured in this study [23].Data are presented as mean ± SD.Highlighted values are significant (P < 0.05).Abbreviations: AD, Alzheimer's disease; BM, Bruch's membrane; ELM, external limiting membrane; GCL, ganglion cell layer; INL, inner nuclear layer; IPL, inner plexiform layer; NFL, nerve fiber layer; OPL, outer plexiform layer; PR1, layer of the inner segment or ellipsoid zone; PR2, layer of the outer segment or interdigitation zone; RPE, retinal pigment epithelium.
between the cohorts from the literature and this study, and also to keep consistent for comparison, within the cohort of this study.Statistical significance in all analyses was set at P < 0.05.

| Comparison of retinal layer thicknesses
We compared global averages of the retinal layer thicknesses, as well as the average total retinal thickness, between the before-diagnosis and after-diagnosis OCT data for each of the 48 subjects.Results are shown in Table 2 and the spread of the data is graphically depicted in Figure 3.
For the global averages across the 48 subjects, 5 of 10 retinal layers (NFL, GCL, IPL, OPL, and RPE) displayed statistically significant (P < 0.05) thinning.The PR2 retinal layer displayed a significant increase in thickness, and the remaining four layers showed either nonsignificant increases (PR1 and BM) or nonsignificant decreases (INL and ELM) in individual retinal layer thicknesses.The total thickness, the sum of all 10 retinal layers, also displayed a significant thinning.
T A B L E 4 Summary of statistical analysis of retinal layer thickness ratios.

Significance
Change likely related to Summary Ratios Significant AD 5 ratios and 5 inverse ratios T A B L E 5 Conclusions from literature, layer, and ratiometric analyses.

Literature analysis Layer Analysis
Ratiometric Analysis The measured changes were also compared to agerelated differences among healthy subjects reported in the literature to determine whether the changes were specific to AD rather than aging alone (Table 3) [23].

Aging healthy subjects
Accounting for the effect of age, we determined which significant changes were primarily AD-related or age-related.Based solely on the results of the agematched controls, 2 of 10 retinal layers (NFL and PR2) demonstrated primarily AD-related changes, with NFL displaying statistically significant thinning (P < 0.05) and PR2 displaying thickening.The significant decrease in thickness of the OPL retinal layer was best explained by being both AD-and age-related, as neither the AD subjects with age factored out or the age-matched healthy subjects displayed independently significant changes.The same phenomenon was seen with total retinal thickness, so its significant thinning is likely both AD-and age-related.Three of 10 retinal layers (GCL, IPL, and RPE), which had significant decreases in agematched healthy subjects but not in AD subjects with age factored out, thus displayed age-related thinning.For the remaining 4 of 10 retinal layers (INL, ELM, PR1, and BM), it was not clear from just retinal layer thickness analysis whether each layer had neither AD-nor agerelated changes, or whether had both AD-and agerelated changes that cancel each other out, as both cases would result in no significant change.

| Comparison of retinal layer thickness ratios
We next compared global averages of the retinal layer thickness ratios between the before-diagnosis and after-diagnosis OCT image volumes for each subject.A summary of results from a paired Student's t-test statistical analysis is seen in Table 4. Complete results from the analysis are shown in Table S1.The compared ratios are seen in Table S2.
In the global averages of layer thickness ratios across the 48 subjects, the 90 ratios can be subdivided into three categories: 28 ratios between two layers that each had a significant thickness change, 48 ratios between one layer that had a significant and one layer that had a nonsignificant thickness change, and 18 ratios between two layers that each had a nonsignificant thickness change.Among the significantjsignificant comparisons, 23 of 28 (82.14%) of ratios themselves were significant, while significantjnonsignificant comparisons had 26 of 48 (54.17%) and nonsignificantjnonsignificant comparisons had 4 of 14 (28.57%) of ratios themselves being significant.In total, 53 of 90 (58.8%) retinal layer thickness ratios were significantly different between the before-diagnosis and after-diagnosis OCT data.
We also analyzed if each ratio and its inverse had a consistent statistical result and found that in 44 of 45 (97.8%) pairs, a ratio and its inverse were either both significant or both nonsignificant.The one pair with a discrepancy involves NFL and IPL, two layers that independently demonstrated significant thinning.
To determine if neither or both AD and age are contributors for the four retinal layer thicknesses with no apparent significant changes (INL, ELM, PR1, and BM), retinal layer thickness ratios were analyzed.In comparing the ratios among these four layers, three layers (ELM, PR1, and BM) each have significant ratios with one another, while INL remained without a significant ratio.From these analyses we were able to draw conclusions regarding the role of AD and age in retinal layer thickness changes, as seen in Table 5.

| DISCUSSION
While progress is being made in understanding the pathophysiological processes of Alzheimer's disease, many biochemical and microstructural changes relating to AD and the emergence of clinical symptoms remain unclear [24].Clinical biomarkers have been used successfully as diagnostic tools to increase our certainty that patients presenting with dementia likely due AD have findings consistent with other AD patients [5].These clinical biomarkers thus aid our understanding of the AD pathophysiological process with respect to identifying a clinical basis for dementia.As the neural retina is an extension of the brain embryologically, anatomically, and physiologically, there is reason to believe there exists retinal biomarkers for AD, fundamentally a neurodegenerative disorder.
The results of this study yielded several interesting findings that contribute to our understanding of how AD induces changes in retinal cell layers.Considering the measurements of global averages of retinal layer thicknesses across the macula, there were significant decreases in half of the retinal layers (NFL, GCL, IPL, OPL, and RPE).Three of these layers (NFL, GCL, and IPL) are largely comprised of retinal ganglion cells, projection neurons that relay visual information to the brain.Retinal ganglion cell layer degeneration in AD is backed by longstanding findings in the literature, which reports that retinal NFL and IPL are significantly thinned in patients with dementia due to AD and in patients with preclinical AD [25].Additionally, some of the layers we found to have significant decreases have been found to demonstrate thinning with other retinal diseases, such as the thinning of the NFL with retinitis pigmentosa [26].Since we excluded subjects from the study that had any concomitant retinal disease, and controlled for age, we believe that the NFL thinning observed in this study is primarily due to AD-related degeneration rather than an effect of other retinal pathologies or age.Similarly, we determined that the IPL and GCL thinning observed in this study is primarily age-related.
Two other layers that were found to be significantly thinned include the OPL of the inner retina and RPE of the outer retina.Thinning of these two layers have been reported in both aging and AMD [27,28].Again, our study excluded subjects with any diagnosis or medical history of AMD, so the global decreases in retinal layer thickness that we identified is likely due to aging, which has reported associations with thinning of the NFL, GCL, IPL, INL, OPL, and RPE, particularly in the seventh decade of life [29][30][31].To isolate changes due to AD alone for our cohort with an average age in the ninth decade, we compared retinal layer thickness changes to values reported in a study of healthy subjects of similar age [23].The significant thinning reported in the literature here was contained in the GCL and IPL layers between the eighth and ninth decades, and the PR1, PR2, RPE, and BM layers between the 9th and 10th decades.Accounting for these normative results, our findings show significant thinning and changes in associated ratios across multiple layers can be more accurately attributed to pathological changes associated with AD in the observed retinal thicknesses and ratios, rather than due to aging alone.Specifically, we found that OPL thinning and total retinal thickness thinning observed in this study is best explained by the combination of AD-and age-related degeneration.
A peculiar finding is the significant thickening of the interdigitation zone (PR2).While this layer thickness has been measured to change in a manner that correlates inversely with glaucoma severity, it has not been described in the literature to increase in thickness with retinal or neuropathological disease processes.Additionally, we determined this thickening to primarily ADrelated.Further investigation over a larger population of subjects will be needed to validate this finding.
The physiological process of aging can also increase certain retinal layer thicknesses, for example, thickening of the BM layer, between the third and seventh decades of life [32].In our study, there was a nonsignificant increase in the BM, which could potentially be indicative of further thickening in the increased age of the study subjects.To rule out potential impacts of physiological age-related changes (associated with thinning of various layers) or undiagnosed pathological AMD (associated with thinning of the OPL and RPE), we compared retinal layer thickness ratios.In a ratiometric analysis, a shared physiological factor that causes thinning in each layer is hypothetically eliminated from the equation when a ratio between the two is taken, thus leaving only mechanistic differences between the thinned layers to explain why their discrepancy may be significant.
In both significantjnonsignificant comparisons and nonsignificantjnonsignificant comparisons, ratiometric analysis uncovered additional significant changes than what would have been identified from comparing individual retinal layer thicknesses alone.Furthermore, the layers that were most involved in significant ratios (IPL, GCL, and NFL) were independently significant, while the layers that were least involved in significant ratios (INL and ELM) were independently nonsignificant.In these ways, ratiometric analysis provided more information than layer thickness measurements alone, proving to be a more sensitive indicator of changes.
An interesting finding from the ratiometric analysis is that the IPL/NFL ratio is significant while the NFL/IPL ratio is nonsignificant.While each layer has retinal thinning that is significant on its own, it is curious that the IPL/NFL ratio is still significant and its inverse is not.In the 44 other ratios, a ratio and its inverse share the same significance.Among ratios where each layer independently had significant changes, most of their ratios are significant.Important to note are the significant ratios where neither layer was independently significant: PR1/ELM, ELM/PR1, BM/ELM, and ELM/BM.These results suggest that the retinal thinning in PR1 and ELM are correlated and that the retinal thinning in BM and ELM are correlated, as the effect of a potential common biological mechanism between a pair of layers is factored out when taking a ratio of the two layers.While these three layers along with INL are the four layers that are not independently significant, all four layers have significant ratios.Furthermore, the correlations between the ratios involving PR1, ELM, and BM in conjunction with the age-matched control results suggest that both AD-and age-related changes affect these layers.Similarly, the lack of involvement of INL with other layers along with age-matched control results suggest neither AD-nor age-related changes significantly affect INL.
The application of age-matched control data and analysis of retinal layer thicknesses and ratios suggest that the changes in NFL and PR2 are primarily ADrelated; changes in GCL, IPL, and RPE are primarily agerelated; changes in OPL, ELM, PR1, BM, and total retinal thickness are both AD-and age-related; and changes in INL are neither AD-nor age-related.
This ratiometric analysis paired with a strong biological understanding of retinal layer thinning could establish OCT image-based biomarkers that may correlate to the pathophysiology of neurodegeneration seen in Alzheimer's disease.On its own, the ratiometric analysis conducted here suggests retinal layers that may be associated with changes in AD, and those that should be further investigated in neurobiological research.The computational and automated image segmentation process we utilized here offers a potential platform to identify correlations between retinal layers.While significance was determined simply for individual layers and ratios of layers, one could perform additional directed mathematical operations including logarithmic and exponential manipulations to combine thickness measurements in different ways.For example, a multiplication approach can be used to amplify the differences within a set of before-diagnosis and after-diagnosis OCT scans.Through these mathematical operations we can identify potential relationships between layers never before reported in the literature.
We recognize in this study that the patient number and image volume number are low, largely because this study involved retrieving image data retrospectively, instead of prospectively.We were also not able to control some of the acquisition variables.We do note that the images collected were part of the clinical standard-ofcare, and we believe there was general consistency in how the image data was collected because of this standard-of-care practice.There may be variation in how data was collected between the before-and afterdiagnosis scans.It should be noted however that each of these scans came from the same ophthalmologist's office in a relatively short time frame.To the best of our knowledge, we believe that the standard operating procedure was very similar both before and after diagnosis, and we did not identify significant differences in the data format or collection parameters.Each set of image data were also collected on the same make and model of commercial OCT system.
A limitation of this study is that we controlled for age with and compared our results to a literature-reported study instead of a retrospectively collected control group.We determined this to be acceptable as the OCT system used in both the literature-reported study and this study was the same commercial ophthalmic OCT system (SPECTRALIS Optical Coherence Tomography, Heidelberg Engineering, Inc.).Additionally, the segmentation software used in both studies was the automated Heidelberg software.While there may be some settings or approaches that differ between the two studies, the main metrics (age and age-related OCT-measured layer thicknesses) were ensured to be as close as possible so that the results could be compared directly.In our comparison we made our best effort to control for age, selecting three healthy cohorts from the literature with age ranges that roughly overlapped with our own study cohorts.While the ages do not match up perfectly, and the time interval between the two OCT sessions did vary, these can be considered to be limitations to the study that can be addressed in a follow-on prospective study in which we can implement more stringent inclusion criteria and acquisition parameters.
Another limitation of this study is the lack of diversity in the demographic characteristics of the subjects.In this study the ratio of female subjects to male subjects is 2.2:1, however, a study with a more even split could help reduce bias in the results that are potentially due to sexrelated differences.Additional studies with separate female and male subject cohorts could help identify these differences.Similarly, a future study could benefit from including subjects from different racial groups, for the current data is highly skewed toward White Caucasian subjects as a result of the general patient population seen in an ophthalmology clinic in central Illinois.It could also be suggested that this study should have age-matched controls of subjects with no AD diagnosis who undergo OCT screening at multiyear intervals; however, as collecting corresponding pairs of OCT volumes on healthy subjects would take multiple years to complete, we utilized and cited a previous study on agerelated OCT retinal changes in healthy subjects, and have incorporated age-matched healthy subject imaging in ongoing future studies [23].
Parameters known or hypothesized to be associated with AD, including the correlative thickness and ratiometric changes we identified from OCT images, can be selected for a machine-learning based algorithm tasked to detect AD during its early stages.Recent work accurately predicted short-term progression of mild cognitive impairment to AD through a machine learning-based algorithm trained on parameters including imaging, CSF, genetic factors, cognitive resilience, and demographics [33].Such studies enable ranking of various biomarkers as predictors, and from these sets of predictors, selection of a small set of minimally correlated features that could be used to accurately predict progression of the disease.If these future machine-learning studies reveal that retinal thickness changes can be classified as a diagnostic biomarker, there is promise that we could achieve earlier detection of Alzheimer's disease through noninvasive OCT eye scans.
An effective noninvasive, label-free ophthalmic imaging approach used for both early detection of AD, as well as interval tracking of progression or regression of the disease following various treatment strategies, has the potential to provide great benefit to not only the increasing numbers of patients in our aging population, but also to those involved in the delivery and management of their care [34].
Results of the three analyses performed in this study (literature, layer, and ratiometric) were compared to explain the etiology of statistically significant (P < .05)changes.Abbreviations: AD, Alzheimer's disease; BM, Bruch's membrane; ELM, external limiting membrane; GCL, ganglion cell layer; INL, inner nuclear layer; IPL, inner plexiform layer; NFL, nerve fiber layer; OPL, outer plexiform layer; PR1, layer of the inner segment or ellipsoid zone; PR2, layer of the outer segment or interdigitation zone; RPE, retinal pigment epithelium.
Subject demographic and age data.
T A B L E 1Note: Data are presented as mean ± SD.Abbreviations: AD, Alzheimer's disease; OCT, optical coherence tomography.