All analyses were performed by the same grader trained on advanced retinal image analyses (J.H.) using ImageJ (software version 2.0.0; National Institutes of Health, Bethesda, MD, USA). First, each MA was identified in the early phase FA image and assigned to the corresponding location in the AO fundus image. AOOCT volumes of the best quality were chosen for further analysis. All AOOCT volumes were rotated to align the retinal pigment epithelium in the B-scan horizontally before assessing the en face view in order to ensure even levels of retinal capillary plexuses. Image analysis was based on a complete review of each AOOCT volume stack in all three dimensions for each MA. Single, preselected AOOCT images were chosen as figures for publication to illustrate MA morphologic criteria of interest. At baseline, all AOOCT volumes were evaluated for the intraretinal position of each MA, which was defined as the retinal layer where the center of the MA was located. At baseline, the number of feeding and/or draining vessels of the MA and their origin from the individual retinal capillary plexuses (superficial, intermediate, and deep capillary plexus) were recorded. At baseline and at every 3-monthly follow-up visit, the shape of the individual MA was classified in the en face plane of the AOOCT volumes as being either saccular, fusiform, or a focal bulge according to the classification scheme proposed by Moore et al.
19 MA intraluminal reflectivity was graded by comparing it to the MA wall reflectivity. It was considered hyperreflective if the intensity of the reflected signal was similar or stronger compared to that of the MA wall or hyporeflective otherwise. In addition, the homogeneity of the hyperreflective signal as well as the presence of well-circumscribed hyperreflective material in an otherwise hyporeflective MA lumen was assessed.
In a second step, the stacks of the follow-up visits were compared to those at baseline in order to investigate specifications regarding a morphologic change of the MA including MA shape, MA luminal appearance, MA division, and MA size.
Due to the lack of a validated three-dimensional segmentation algorithm for AOOCT volume stacks, precise MA volume measurements cannot be performed at present. For the purpose of this explorative longitudinal study, change in size was measured in the B-scan view of the stacks with respect to the extension of the MA over the individual retinal layers, as well as in the en face view with respect to the largest visible diameter of the MA. In both views, the outer MA wall borders were considered for all measurements. First, the scan displaying the maximum size of the MA was identified in both views in the full AOOCT volume. In order to assess the variability in respect to the chosen B-scan location, the size of the MAs was measured in seven adjacent scans (three adjacent scans in both directions from the “maximum size scan,” respectively). The size of the MA was then defined as the mean of these measurements, and the variability was defined as the standard deviation of this measurement. Microaneurysm size is presented in micrometers (μm) for all results. Images were scaled using the previously calibrated axial pixel size of the instrument (1 pixel corresponds to 3 μm, assuming a refractive index of 1.41 of the retina) and the axial eye length (AL) information of all study patients. For lateral scaling the scanning angle was calibrated using a model eye and a calibration target as retina. With a scanning angle of 2°, the conversion from angle to distance on the retina (x) was performed using following equation: x = 2* ((AL − AD)/1.33)* tan (1), where AL is the axial eye length, AD is the anterior chamber depth, and 1.33 is the refractive index of water. As an example, for an eye length of 24 mm, AD = 3.5 mm, two degrees of scanning angle correspond to 540 μm. This distance is sampled with 400 pixels, which results in 1 pixel corresponding to 1.35 μm. All these two-dimensional measurements were performed by the same grader, and the evaluation of intergrader variability was forgone at this stage, considering the lack of a robust three-dimensional AOOCT segmentation algorithm.