Skip to main content
Dryad logo

A practical approach to functional optical coherence tomography shows abnormal retinal responses in Alzheimer’s disease

Citation

Bissig, David; Zhou, Clarice; Le, Vy; Bernard, Jacqueline (2020), A practical approach to functional optical coherence tomography shows abnormal retinal responses in Alzheimer’s disease, Dryad, Dataset, https://doi.org/10.5061/dryad.msbcc2ftc

Abstract

Spectral-domain optical coherence tomography (SD-OCT) is an accessible clinical tool for measuring structural changes to the retina, and increasingly as a biomarker for brain-predominant neurodegenerative diseases like Alzheimer’s. Information about retinal function can also be extracted from OCT images, but is under-studied, with literature examples often employing challenging protocols or requiring specialized hardware. The first goal of this study was to verify that functional retinal imaging was feasible with a commercially-available SD-OCT device and a clinically practical protocol. Inspired by methods from other functional imaging modalities, we acquired images while repeatedly cycling lights on and off, and spatially normalized retinas to facilitate intra- and inter-individual analyses. In eight healthy young adults, light-dependent increases in reflectivity were easily demonstrated at photoreceptor inner and outer segments, changing by ~7% in bright light and ~3% in dim light. Bright light elicited a subtle (~2%) but consistent light-dependent decrease in reflectivity through much of the rest of the retina, including the avascular outer nuclear layer (ONL). We speculated that some of these changes are influenced by glial function – as through water management – a topic of high interest in neurodegenerative diseases that may involve the glymphatic system. Functional abnormalities in patients with antibodies against aquaporin-4 (n=3) supported this interpretation. We next compared patients with early-onset Alzheimer’s disease (n=14) to age-matched controls (n=14), revealing that patients had a relatively exaggerated light-induced change in ONL reflectivity (p<0.05). Because these measurements can be obtained within thirty minutes, regular use in research and limited clinical settings is feasible.

Methods

COLLECTION:

  Images were collected on a Heidelberg Engineering SPECTRALIS OCT. In Experiments 1, 3, and 4, The SPECTRALIS Slice Planning Tool was used to select a single slice of the retina per subject, which was automatically re-selected for each acquisition. That single slice spanned 30° eccentricity, including and angled through the (para-)fovea and the center of the optic disc, generating the cross-sectional view seen in Figure 1 of the manuscript. Automatic Real Time (ART) averaging was set to the maximum (100). Images were collected after > 20 seconds of exposure to light ("_LIGHT"), or after 2 minutes of darkness ("_DARK"). We cycled through light and dark repeatedly, collecting up to fifteen images per participant per condition. In Experiment 2, two perpendicular slices were collected. Both spanned 30° eccentricity. Due to software limitations, neither the horizontally-oriented nor vertically-oriented slice could be tilted to cross the fovea and center of the optic disc. However, the horizontally-oriented slice approximates the slice position used in other experiments.  

PROCESSING:

  Raw data were extracted from .sdb files using ImageJ (v.1.47)(Import>>RAW>>16-bit unsigned, with little-endian byte order). Retinas were cropped from raw reflectivity maps, resampled from the native image resolution to isometric pixels (3.89 μm × 3.89 μm). In experiments using a bright stimulation light, these images were saved in the ANALYZE format, named according to the convention "[SubjectID]_DARK" or  "[SubjectID]_LIGHT" and are readily viewed by dragging and dropping into ImageJ. For young adults (SubjectID starts with "YA"), additional images were collected under a dim stimulation light, denoted by the suffix "-dim". Pixel values in these images are directly proportional to reflectivity.

  Next, the approximate position of the following structures was manually marked: the border between the vitreous and the retina, the border between the retinal nerve fiber layer (RNFL) and the ganglion cell layer (GCL), the border between the inner plexiform layer (IPL) and the inner nuclear layer (INL), the border between the outer plexiform layer (OPL) and the typically hyporeflective band containing both the Henle fiber layer (HFL) and the outer nuclear layer (ONL), the outer limiting membrane (OLM; also called the external limiting membrane), and the RPE. Raw and manually-marked data were saved in the ANALYZE format for later processing using a custom R (v.3.3.3) script. In each subject's folder, ANALYZE images with the "_MARKED" suffix denote those with manually-labeled layers. The R script and an analysis example are provided in the folder "_PROCESSING_EXAMPLE01".

  The following steps are performed by the custom R script: First, the apparent angle of the retina was measured for later use. This value reports on entrance position of the OCT machine’s infrared beam, which impacts the angle at which it encounters the retina. For the Henle fiber layer in particular, even small changes in retinal angle (in the “bright” Experiment 1 data for these young adults, 25th percentile, median, and 75th percentile were respectively -0.83°, 0.90° and 2.72°, where positive values reflect a counter-clockwise rotation from horizontal), can significantly impact reflectivity. Next, retinas were linearized based on the contours of the RPE. These ANALYZE images are stored with the prefix "_flat_". In these flattened retinas, the position of aforementioned layers/structures was automatically refined based on local signal features. Subsequent steps were applied to a large span of the retina that was maximally illuminated when the stimulus light was “on”: Measuring along the RPE, this spanned 500 μm to 2750 μm away from the center of the fovea, towards the optic nerve head. Since layer thickness varies between subjects, and within each subject as a function of distance from the fovea, remaining images were spatially normalized to a common template. At each distance from the fovea, we sampled signal at a specific number of points spaced evenly between the aforementioned anatomical borders: (a) 8 points between the vitreous/RNFL and the RNFL/GCL borders, (b) 22 points between the RNFL/GCL border and the IPL/INL border, (c) 16 points between the IPL/INL border and the OPL/HFL border, (d) 16 points between the OPL/HFL border and the OLM, and (e) 18 points between the OLM and the RPE. Note that where a layer is thick (e.g., RNFL near the optic disc), those points are spaced farther apart than where the layer in thin (e.g., RNFL near the fovea). By dynamically over- or under-sampling based on layer thickness, a universal map of signal intensities as a function of location is generated. This map is available in ANALYZE format with the "_flat-normed_" prefix. Finally, this map is averaged to create a profile of reflectivity as a function of %Depth into the retina for each image (with the retina-vitreous border at 0%Depth, and the RPE at 100%Depth, and intervals of 1.25%Depth). For completeness, five points are evenly sampled between 4 μm and 12 μm interior to retina (into the vitreous) and 4 μm and 24 μm exterior to the RPE (into the choriocapillaris) and placed at 1.25% intervals on this %Depth scale. These profiles are stored in .xlsx files in each folder.

  Profile data are also stored in PROFILE_DATA_for_Dryad.xlsx, alongside participant age ranges and other data. Statistical comparisons were performed by loading the tab-delimated version of the table (PROFILE_DATA_for_Dryad.txt) into R. These analyses scripts are not included in this repository, but are available upon request. In those final steps of the analysis, reflectivities were normalized: In each profile of reflectivity as a function of %Depth into the retina, reflectivites were ratio-normalized to the RPE (i.e., each value was divided by the reflectivity at 100%Depth, and multiplied by 100). It is also within these final R scripts that we were able to use the apparent angle of the retina as a covariate, accounting for the layer-specific impact of beam angle on reflectivity.

 

Usage Notes

  Missing and Excluded Data FOR EXPERIMENT 1: No subjects had gross signs of retinal pathology on infrared fundus images, nor on OCT images. Images were individually inspected to verify proper acquisition, adequate signal-to-noise, and the absence of imaging artifacts. In total, 16 of 320 images were excluded. Each subject had at least 12 “bright” images in both lights “on” and lights “off” conditions, and at least 4 “dim” images in both “on” and “off” conditions.

  Missing and Excluded Data FOR EXPERIMENT 2: In total, 9 of 60 images were excluded (five “off” and four “on” images).

  Missing and Excluded Data FOR EXPERIMENT 3: Only one image was excluded, an “on” image from the third patient. The first patient had extensive degeneration of the GCL, and data assigned to the GCL area by the spatial normalization program were therefore censored.

  Missing and Excluded Data FOR EXPERIMENT 4: One Alzheimer’s patient had a sub-RPE deposit (presumably drusen) near the fovea, but the rest of the retinal strip had a benign appearance, visual fields were full, and Snellen acuity was 20/30-2. Reflectivity profiles (in the subject-specific .xlsx file, as well as PROFILE_DATA_for_Dryad.xlsx) were calculated using the span from 1000 to 2750 μm from the fovea, bypassing the deposit. Although a different participant had an abnormally thin RPE directly at the fovea (within 150 μm of its center), all data were retained due to the otherwise benign appearance of the retina, normal visual fields, and 20/30-2 Snellen acuity. Again, all images were reviewed for quality, and only five were excluded (each in a different participant), leaving 275 images.

  Exclusions due to quality or artifact (where present) are listed in a plain .txt file in each participant's folder.