Longitudinal normative OCT retinal thickness data for wild-type mice, and characterization of changes in the 3×Tg-AD mice model of Alzheimer's disease

Ethics statement

This study was approved by the Animal Welfare Committee of the Coimbra Institute for Clinical and Biomedical Research (iCBR), Faculty of Medicine, University of Coimbra. All procedures involving mice were conducted as per the Association for Research in Vision and Ophthalmology statement for animal use, and in agreement with the European Community Directive Guidelines for the care and use of nonhuman animals for scientific purposes (2010/63/EU), transposed into the Portuguese law in 2013 (DL113/2013).

Mouse characterization

Fifty-seven male mice from each strain, C57BL6/129S and 3×Tg-AD, were used, and both eyes imaged at the ages of one, two, three and four-months-old. Mice from both groups were enrolled in the study within nine months as they became available. All mice were housed and maintained at the vivarium of the Coimbra Institute for Clinical and Biomedical Research (iCBR), Faculty of Medicine, University of Coimbra, and were on a 12-h light/dark cycle with free access to both food and water.

Experimental setup

Concerning OCT imaging preparation, mice were anaesthetized using a mixture of 80 mg/kg of ketamine (Nimatek; Dechra) and 5 mg/kg of xylazine (Sedaxylan; Dechra). The pupils were dilated using a solution of 0.5% tropicamide (Tropicil; Edol) and 2.5% phenylephrine (Davinefrina; Dávi). Additionally, oxibuprocaine (Anestocil; Edol), a local anaesthetic, was used. Eyes were regularly lubricated using eye drops (1% carmellose (Celluvisc; Allergan)).

All retinas were imaged by a Micron IV OCT System (Phoenix Technology Group, Pleasanton, CA, USA). It creates a volume per acquisition composed of 512 B-scans, each with 512 A-scans of 1024 pixels in length; B-scans are saved as a non-compressed TIFF file image. The system presents an imaging depth of 1.4 mm and an axial resolution of 3 μm, as determined by the bandwidth and central wavelength, respectively 160 and 830 nm, of the superluminescent diode used. All scans were taken by the same operator (JM) in the same retinal region using the optic disc as a landmark; centered horizontally with the optic disc, and vertically above it (Figure 1A).

Segmentation

In this work, the OCT data is used for the segmentation of eight distinct layers, allowing the determination of the following: retinal nerve fibre layer and ganglion cell layer complex (RNFL-GCL), inner plexiform layer (IPL), inner nuclear layer (INL), outer plexiform layer (OPL), outer nuclear layer (ONL), the inner segment of the photoreceptors (IS), the outer segments of the photoreceptors (OS), and the retinal pigmented epithelium (RPE).

The segmentation is achieved using a fully convolutional neural network (FCNN), following a U-type architecture [7], which is composed of two main parts, an encoding path and a decoding path. Each level reduces the feature maps' size in the encoding path, effectively computing increasingly global features. The decoding path takes these global features and expands them to classify individual pixels in one of the layers considered in this study. Furthermore, these neural networks can have short-circuit connections [7]. These pathways connect encoding and decoding levels allowing the use of local features, obtained along the encoding path, to provide finer grain information and produce more accurate pixel classification predictions. The neural network model used, with the characteristics detailed above, is disclosed in [8].

In brief, while the network was designed to classify pixels into one of the retina's eight layers, the main objective is to ensure the proper discrimination at these layers' interfaces. The ground-truth interface samples were obtained based on segmented volumes (Figure 1B).

Because of the limited training set and to eliminate the possibility of the network to learn from the usual location of retinal layers within B-scans, the training set was augmented in two ways: by mirroring each B-scan horizontally, and by modulating the location of the retina across the B-scan image. The latter was achieved by circularly-shifting each A-scan across the B-scan by an amount given by a sinusoid which parameters, amplitude, frequency and phase, were randomised per B-scan. The segmentation provided by the neural network was then validated using manual segmentation examples provided by two graders (HF and Raquel Boia).

Thickness

The segmentation of the entire volume (Figure 1B), is achieved by segmentation of each of the 512 B-scans. All thickness maps, per retinal layer and the TRT, are computed as the distance between respective boundaries and defined by 512×512 thickness values.

Due to the nasal-temporal asymmetry found between left and right eyes, all left-eyes were mirrored to match the right ones to ease comparisons. As such, left-hand-side/right-hand-side to thickness maps corresponds to the temporal/nasal region, respectively.

Besides the possibility of disclosing the average retinal thickness for the entire imaged region of the eye, thickness maps of 3×3 values were computed as the average thickness of 170×170 thickness values of the original 512×512 thickness maps (Figure 1A) after cropping it to 510×510 pixels so that all blocks are of equal size.

Average thickness is presented both for the entire imaged area and for 3×3 blocks. The former is the standard measure disclosed in the literature, while the latter provides some data regarding thickness variation over the scanned area.

Statistical analysis

Individual thickness values, each of the 512×512 values of each thickness map, were assessed for their segmentation quality based on multiple criteria including image quality, segmentation (boundary) consistency, and computed thickness values distribution. Those not fulfilling one or more of the established criteria were individually excluded from all analysis.

For establishing the reference thickness maps, only blocks with 90%, and above, of the 170×170 thickness values fulfilling the above criteria were considered. Using these criteria, the percentage of blocks removed was 9.8%, 4.8%, 10.7%, 1.2%, 0.7%, 1.0%, 0.7%, 0.7% and 0.9%, respectively for blocks one to nine, as in Figure 1A.

The Kolmogorov-Smirnov normality test was applied to each group, finding that the large majority of distributions do not reject the null hypothesis that the distributions are normal. Hence, the t-test was used to determine significant statistical differences between WT and 3×Tg-AD groups. Significance levels of 5%, 1%, and 0.1% were used in this work.

For the analysis of the thickness data over time, at the ages of one, two, three, and four months, the repeated measures ANOVA (RANOVA) was performed. The same analysis was performed to study differences between each block in the 3×3 thickness map. The multiple comparisons corrections performed in this work used the Tukey test.

All data processing and statistical treatment were performed using Matlab R2020a (The MathWorks Inc., Natick, MA, USA).

Weight

As shown in Table 2, weight consistently increases over the reported period for both WT and 3×Tg-AD mice. Furthermore, the highest increase in weight occurs between one and two-months-old, as the mice reach their maturity at the ages of six to eight weeks, after which growth slows down.

Comparing weight distributions for both groups did not find statistically significant differences between groups at any of the reported ages.

Thickness maps

Detailed average thickness maps of 510×510 values were computed for the right and left-eyes per each retinal layer and the TRT. The latter is shown in pseudocolor in Figure 2, for WT and 3×Tg-AD groups, at each time point.

Besides the apparent nasal-temporal asymmetry, thickness values are within the same range for both eyes of the same mice group at the same time point. Indeed, a thicker nasal side was found with statistically significant differences to the temporal side for all layers/layer-aggregates, except for the IPL, in three or more time points for the right-eye of WT mice.

Over time, the thickness difference between the ages of one and two-months-old stands out as notoriously greater than the thickness difference between the remaining consecutive ages. Nevertheless, while the TRT showed a consistent decrease over time, individual layer thickness behavior differs, as shown ahead in this section.

Mean thickness

Tables 3, 4 show the average (and standard deviation) thickness values, per retinal layer, for WT and 3×Tg-AD mice, combining the data from both eyes, for the entire imaged area (Figure 1A). For detailed data, please refer to Supplementary Tables 1–4.

Comparing right to left eyes within groups found no statistically significant differences. On the other hand, as illustrated in Tables 3, 4, the comparison between groups shows consistent retinal layer differences. Overall, the 3×Tg-AD group's eyes present a decreased layer thickness except for the RNFL-GCL complex, where the 3×Tg-AD group present an increased thickness, and the ONL, where thickness values match one another.

Statistically significant differences between groups can be found in Table 5 for both eyes. While the highest statistical differences were found for the IPL, INL, OPL, IS, RPE and the TRT, no differences were found for the ONL at any of the significance levels considered in this work. The RNFL-GCL complex showed statistically significant differences starting at the age of two-months-old on both eyes. On the other hand, the OS layer presents statistically significant differences at the age of one and four-months-old for both eyes.

Presented data also shows the consistent decrease in thickness for the INL and IPL, respectively of 20% and 10% across the timespan covered, which was partially counterbalanced by the consistent increase of the RPE layer thickness of about 15% for the same period.

Using the repeated measures ANOVA (RANOVA), the thickness distributions were compared across the four-time-points for each group/eye combinations. Only eyes successfully imaged and processed at all four-time-points (WT (N=37) and 3×Tg-AD (N=35)) were considered for this particular analysis. Achieved results show thickness to be dependent on age for the period covered in this study. Pairwise comparisons are illustrated in Figures 3, 4 for the TRT, respectively for the WT and 3×Tg-AD groups. Statistically significant differences were found between the first and remaining time points for all eyes. Also, except for the left-eye of the 3×Tg-AD group, all remaining eye present statistically significant differences between the four and all remaining time points. For detailed data, please refer to Supplementary Figures 1–16.

Normative thickness map

In addition to the above mean thickness values, we also present the average thickness (and standard deviation) for each layer, for each of the blocks (as in Figure 1A), for OD and OS eyes of WT mice, Supplementary Tables 5–8. Likewise, based on the asymmetry found between right and left eyes, thickness values obtained by combining data from both eyes are also shown in Supplementary Tables 5–8. It is noteworthy that the TRT thickness presented in these Tables does not need to exactly match the sum of the thickness of the remaining layers, as the block selection process detailed in section 2.6 is done independently for each layer.

Because most thickness values present a normal distribution for all the nine blocks, all retinal layers at hand in this study and the four-time-points, average and standard deviation values for each block are enough to establish the normative thickness distribution. Furthermore, it allows computing the probability (p-value) of a particular thickness value to be out of the normal range. Section 3.5 will explore further this subject.

The normative data presented in Supplementary Tables 5–8 demonstrates the distance dependence of retinal thickness to the optic disc. In general, thickness decreases as the distance to the optic disc increases. However, the OPL, in addition to the IS, and OS layers, presents a homogenous thickness across the entire imaged region.

Even though the average thickness values for the entire imaged region have shown evident thickness differences over time, averaging smaller regions allows for a better picture of the retina's ongoing changes. For instance, while the average OPL thickness did not show any difference in thickness over time, its detailed analysis shows a decrease in thickness in regions further away from the optic disc counterbalanced by the increase in thickness in regions closer to it.

Use of normative data

In the previous section, the normative thickness map distribution was established for each of the eight retinal layers assessed in this work and the TRT. In the current section, we use those normative thickness maps to show their usefulness in evaluating individual retinas.

Following the acquisition of OCT data and the segmentation of the layers of interest, computed thickness values for the 3×3 thickness map can be matched with the respective normative data to determine either the increased or decreased thickness and the probability (p-value) of the normative distribution to generate a more extreme value than the given thickness, for each of the considered four-time-points. The p-value is determined by the established normal thickness distribution for each block based on mean and standard deviation values. It relates the considered thickness with the normative distribution, being one (p=1.00) for the mean value and decreasing as values get further away from it towards the normal distribution tails.

Two examples, one for WT and one for the 3×Tg-AD, at the age of two-months-old, can be seen in Figure 5, respectively for the INL and the OPL.

For the ease of interpretation, blue and red colors represent thickness values below and above the average thickness; color intensity increases with the probability of respective thickness values being extreme.

The intense blue color of the INL thickness of the 3×Tg-AD mice shown in Figure 5 indicates the overall decrease in thickness compared to the established normative data. Indeed, five out of the nine values present p-values under 0.1, meaning that the probability of randomly generating that or a more extreme thickness value for the normative distribution of the thickness is below 10%. On the other hand, for the WT thickness map, one can find both blueish and reddish colors, meaning that the thickness is around the average, respectively below and above it, with the light colors indicating these are close to the average; six of the nine values present a p-value above 0.6.

On the other hand, both maps for OPL present lower color intensities than the previous comparison, meaning that the two eyes present thickness values closer to the normal range for this layer, as conveyed by the p-values reported. This is in agreement with the above findings for this retinal layer, for which no statistically significant differences were found between groups.