Comparison of multicolor scanning laser ophthalmoscopy and optical coherence tomography angiography for detection of microaneurysms in diabetic retinopathy

This study aimed to evaluate the usefulness of multicolor (MC) scanning laser ophthalmoscopy (MC-SLO) in detecting microaneurysm (MA) in eyes with diabetic retinopathy (DR). This was a retrospective cross-sectional study. Eyes with DR underwent fluorescein angiography (FA), MC-SLO, optical coherence tomography angiography (OCTA), and color fundus photography (CFP) were analyzed. The foveal region was cut in an 6 × 6 mm image and the number of MA in each image was counted by retina specialists to determine the sensitivity and positive predictive value. FA results were used as the ground standard. MAs were classified as those with early, late, or no dye leakage based on FA images. Fifty-four eyes of 35 patients with an average age of 64.5 ± 1.24 years were included. The sensitivity of MA detection was 37.3%, 15.3%, and 4.12% in MC-SLO, OCTA, and CFP, respectively (P < 0.01 in each pair).The positive predictive value was 66.4%, 46.4%, and 27.6% in MC, OCTA, and CFP, respectively (P < 0.01 in each pair). Sensitivity for MAs with early leakage was 36.4% in MC-SLO, which was significantly higher than 4.02% in OCTA. MC-SLO was more useful in detecting MA in eyes with DR than OCTA.


Scientific Reports
| (2021) 11:17017 | https://doi.org/10.1038/s41598-021-96371-y www.nature.com/scientificreports/ images are created for each layer, the image merged by these three images could reveal the retinal structure more clearly than a normal color fundus photograph (CFP). On the basis of these characteristics, it has already been reported that MC-SLO is superior to CFP in detecting retinal diseases, including epiretinal membrane and geographic atrophy [23][24][25][26][27] . In our preliminary observation, we found that MAs showed a characteristic morphology in MC-SLO. Additionally, because normal retinal capillaries do not appear in MC as in OCTA, MA can be identified more easily and has high specificity in MC-SLO. Therefore, this study aimed to evaluate the usefulness of MC-SLO in detecting MA in DR compared with that of OCTA and CFP. The study found that MC-SLO had superior detection rate of MA compared with the two other methods but could also detect clinically significant MA.

Results
Subjects. Demographics of the participants are shown in Table 1. The study enrolled 54 eyes of 35 cases (men, 37 eyes of 25 cases; women, 17 eyes of 10 cases). The mean age was 64.5 ± 1.24 years. The mean best corrected visual acuity (logMAR) was 0.239 ± 0.043. The mean refractive error and axial length were − 1.24 ± 0.31 diopter and 24.0 ± 0.145 mm, respectively. There were 33 phakic and 21 pseudophakic eyes. Eleven cases (31.4%) had a history of smoking. Mean HbA1c was 7.36 ± 0.18%. We excluded cases that underwent any treatment for DME within one month of this study; however, 23 eyes had a treatment history, including 12 eyes with panretinal photocoagulation and 9 eyes with anti-VEGF therapy.

Sensitivity and positive predictive value for MAs in CFP, OCTA, and MC-SLO. We detected 1244
MAs by FA, which was the ground truth for MA detection in this study, in all cases. The number of objects that we judged as MAs included real MAs and MA-like objects that could not be confirmed as true MAs in FA images (i.e., objects that were mistakenly judged as MAs), was 117 in CFP, 304 in OCTA, and 642 in MC-SLO. The number of real MAs confirmed by FA images was 56 in CFP, 160 in OCTA, and 444 in MC-SLO.
The mean number of MAs we detected in FA images was 23.04 ± 3.53. The mean number of objects we judged as MAs was 2.17 ± 0.35 in CFP, 5.63 ± 0.56 in OCTA, and 11.9 ± 1.70 in MC-SLO. In addition, the mean number of real MAs confirmed by FA images was 1.04 ± 0.203 in CFP, 2.96 ± 0.35 in OCTA, and 8.22 ± 1.29 in MC-SLO. Therefore, the sensitivity was found to be 4.12 ± 0.82% in CFP, 15.3 ± 1.63% in OCTA, and 37.3 ± 2.41% in MC-SLO (Fig. 1A). The positive predictive value was 27.6 ± 4.68%, 46.4 ± 3.81%, and 66.4 ± 3.40% in CFP, OCTA, and MC-SLO, respectively (Fig. 1B). As a result, the sensitivity and positive predictive value of MC-SLO were significantly higher than those of OCTA and CFP (P < 0.01, Steel-Dwass test, Fig. 1).

Comparison of deep layer and full-thickness image in OCTA .
The sensitivity of MA detection in deep layer images was 5.80 ± 0.96%, and that in full-thickness images was 16.6 ± 2.26% ( Supplementary Figure S1A). The positive predictive value of MA detection in deep layer images was 31.5 ± 5.65%, and that in fullthickness images was 47.7 ± 4.96% (Supplementary Figure S1B). Both sensitivity and positive predictive value were significantly higher in the full-thickness images than in deep layer images (P < 0.01, Wilcoxon test).  Fig. 4).

Discussion
The present results showed that MC-SLO had significantly greater detection sensitivity and positive predictive value of MAs than OCTA in the 6 × 6-mm image area. The MC-SLO image has a digital depth resolution of 3.5 μm/pixel and one scan interval of 14 μm 27 . In contrast, the OCTA 6 × 6-mm scan has a horizontal direction of 11.719 μm and vertical direction of 23.438 μm, which was comparable with MC-SLO. Thus, the difference in the scan interval between machines did not seem to be a decisive factor for the current results.
In terms of appearance, MC-SLO showed the characteristic findings of MA. As in earlier studies 28 , most MAs had a characteristic finding, such as the central green dot with peripheral red color in MC-SLO. This feature made it easy to detect MA and distinguish it from retinal hemorrhages. Hemorrhagic dots showed more reddish color and less green color. According to pathological studies, MAs have a diameter of 50-100 μm [29][30][31] , and proliferation and degeneration of vascular endothelial cells are present in MAs [32][33][34] .
Fibrotic connective tissue, including thickened blood vessel wall, appears green in MC-SLO. Moreover, the red blood cell components in MA appear red 28 . Thus, it is feasible that the central green color of MAs reflects fibrotic changes and thickened vessel walls in MC-SLO. The second one was related to the mechanism for creating   In contrast, OCTA had a low MA detection rate with 12.3%, which seemed lower than that in earlier studies, from 41 to 62% in 3 × 3-mm OCTA image [17][18][19][20] . It is possibly because a 6 × 6-mm image was used in this study, which is more commonly used than 3 × 3-mm image in clinical practice. The interval space of 6 × 6-mm image in OCTA is 11.719 μm in lateral scan, and that of 3 × 3 mm image is 5.859 μm. Similarly, the horizontal scan interval was 23.438 μm in 6 × 6-mm image and that of 3 × 3-mm image 20 was 11.719 μm. The larger scan mesh made the detection rate lower. MC-SLO is not affected by that factor. Another reason was the difference in the analysis between this study and a previous study. In previous studies, the detection rate of MA was examined by superimposing (merging) FA and OCTA images 17,19,20,35 . However, our raters evaluated the MAs in OCTA image separately, and the other rater calculated the detection rate after the primary analysis. Indeed, Carlo et al. studied the detection rate of MAs in OCTA with a method similar to ours 36 . They showed that the detection rate of microvascular abnormalities was 26.1%, which was comparable with our result.  www.nature.com/scientificreports/ Pathological research has shown that MA is a dilation of capillaries, predominantly in the central retina and mostly originating from the deep capillary plexus 33,34 . Similarly, using OCT B-scans, 67.8% and 80.3% of MAs were observed to be localized in the INL in this study and in Horii et al. 's study 37 , respectively (Supplementary  Table S2). However, Parrulli et al. reported that Topcon OCTA, which is similar to the method used in the present study, detected many MAs (57%) in the deep (IPL-OPL) images but also detected 43% of MAs in the superficial (ILM-IPL) images, although there were differences in the results between the devices 38 . They also showed that more MAs were detectable when the two layers were combined. As shown in Supplementary Figure S2, we were able to detect more MAs in the full-thickness OCTA images than in the deep layer OCTA images. Thus, fullthickness OCTA images were used in the present study. The difference between the layers analyzed in the studies could affect the variation in the results.
Theoretically, an OCTA image is monochromatic and constructed by the movement of red blood cells. Thus, it could show not only MAs but also the surrounding capillaries. This feature might make it difficult to distinguish MAs from surrounding capillaries (representative images are shown in Supplementary Figure S3).
Even when normal capillaries bend or run vertically, it may be indistinguishable from MA. However, MAs could be visualized without showing normal capillaries in MC-SLO. Thus, the sensitivity and positive predictive value may be better in MC-SLO than in OCTA. Another advantage of MC-SLO is its superiority to detect clinically important MAs. Photocoagulation of MAs is performed when the MAs are the major cause of retinal edema 14 . Therefore, we defined MAs with early and late leakage by the findings in FA images and examined the sensitivity in MC-SLO and OCTA. First, the retinal thickness at the MA with early leakage was significantly greater than that of MAs with late leakage. Thus, MAs with early dye leakage are more important for the treatment of retinal edema than those with late dye leakage.
Interestingly, MAs with early dye leakage were found to be less detectable than MAs with late dye leakage in OCTA images. Further examination revealed that MA with early dye leakage was more signal-less than MA with late dye leakage in OCTA. There are two possible reasons as follows: first, in an MA with early dye leakage, the movement of red blood cells would be slow inside the MA, and thus, it may not be displayed as a blood flow signal at the current OCTA scan rate.
Nakao et al. studied the appearance of MA in AO-SLO and OCTA. They reported that MA with turbulence of blood cells was detectable in OCTA 35 . Thus, leaking pattern is thought to be diffuse and multidirectional in the MA with active leakage. These features might slow down blood flow, resulting in fewer signals of the MAs in OCTA. Second, a majority of the MAs with early leakage are associated with retinal edema. Retinal edema may prevent detection of MAs 37,39 . In contrast, the detection rate of MAs with early leakage was not decreased in MC-SLO. The reason is that MC-SLO captures only the "morphology" of MAs and is not affected by blood flow as OCTA.
There are several limitations of this study. First, this is a retrospective study with a relatively small number of cases, and the surveyed area is limited to 6 × 6 mm. Second, we evaluated images by objective methods as described. Although analyses were performed by experienced examiners, this might affect the results. It is necessary to be cautious of these limitations when interpreting and generalizing the present findings.
Third, this study included phakic and pseudophakic eyes. Intermediate opacities, including cataracts, are known to affect image quality to some extent [40][41][42] . Therefore, we excluded eyes with poor image quality owing to intermediate opacity, but the condition of the lens may have possibly affected the results. Finally, Arrigo et al. reported that 20% of MAs appear red dots in MC-SLO and these MAs correspond to type 1 MA, which are MAs with normal vascular endothelium and no pericytes, but with extensive accumulation of monocyte and polymorphonuclear cells in the lumen as per Stitt et al. 's histological MA classification 28,34 . We started this study with the understanding that MAs would be depicted with a green center in MC-SLO. Since the method of analysis and the background of the patients in our study differed from those in the paper by Arrigo et al. (mean age: 55 years in this study vs. 64.5 years in Arrigo et al. 's study, mean logMAR visual acuity: 0.6 in this study vs. 0.2 in Arrigo et al. 's study), the percentage of MAs that appear red dotsin our patients is unclear. However, it is possible that type 1 MAs that appear red were not included.
In conclusion, MC-SLO has higher sensitivity and positive predictive value for MA detection than OCTA. Especially for MAs with active leakage, which is clinically important for DME treatment, MC-SLO could detect MAs better than OCTA. Thus, it is suggested to use MC-SLO for managing eyes with diabetic retinopathy. FA remains to be the gold standard for MA detection; however, due to its invasiveness, MC-SLO is useful in cases where performing FA is difficult or when repeated evaluations are needed, such as for the evaluation of the treatment effect for DME.

Methods
Ethics statement. All procedures used in this study confirmed and complied with the tenets of the Declaration of Helsinki and were approved by the Ethics Committee of Kagoshima University Hospital. The committee considered that a written informed consent was not necessary due to the retrospective nature of the study.
Study design, subjects, and examination method. This was a retrospective cross-sectional study.
Consecutive patients with DR who visited Kagoshima University Hospital between January 2016 and May 2019 and underwent FA, MC-SLO (imaging range, 30°), OCTA (6 × 6-mm map, Triton, Topcon, Japan), and CFP (Triton) were included. Ocular examination was performed within 1 week. Eyes that received DME treatment within one month of the aforementioned examinations and eyes without clear images due to media opacities (such as cataract, which is known to affect the quality of fundus images [40][41][42] were excluded from analysis. FA and MC-SLO images were taken using a Spectralis HRA (Heidelberg Engineering, Heidelberg, Germany) with at least 30 averaged images for MC-SLO. OCTA images were obtained with a central wavelength of 1050 nm, an www.nature.com/scientificreports/ acquisition speed of 100,000 A-scans/s, an axial resolution of 7 μm, and a transverse resolution of 20 μm. Scans were cubes of a 6 × 6 mm dimension with each cube consisting of 320 clusters of 4 repeated B-scans centered on the fovea. The captured images were extracted as TIFF images of the highest quality. All subsequent analyses were performed on the same computer.
Sensitivity and positive predictive value of MA detection in CFP, OCTA, and MC. The images of CFP and MC were cut in the same size as the image of 6 × 6 mm OCTA using Photoshop (Adobe Systems Inc., San Jose, USA). The ground truth of the presence of MA was determined using FA. On the basis of the study by Schreur et al. 19 MAs in OCTA images were defined as hyporeflective, moderate, or hyperreflective spots with various morphologic patterns, including fusiform, saccular, curved, and rarely coiled shapes. MAs in MC-SLO images were also defined as green dot and green dot with peripheral red color partly based on the study by Arrigo et al. 28 . The presence of MAs in each image was determined by two examiners (T.S. and H.T.). Another independent evaluator (H.S.) calculated the sensitivity and positive predictive value based on the following formulas. True MA was determined using FA and used as the ground truth (Fig. 6).

Differences in retinal thickness between MAs with early and late leakage.
In FA, MAs with early dye leakage are supposed to leak more dye than those with late leakage. It is possible that the former is involved in the pathology of DME more significantly than the latter. To assess this hypothesis, we compared the retinal thickness around MAs with early, late, and no dye leakage.
To distinguish MAs with early dye leakage, we performed an evaluation based on the following criteria ( Fig. 7): 1. The size of FA leakage at 30 s was defined as the basic leakage size. 2. The size of FA leakage at 105 ± 15 s was defined as the early leakage size. 3. The size of FA leakage at 270 ± 30 s was defined as the late leakage size.
MA with "early dye leakage" had an early leakage size more than three times larger than the basic leakage size. MA with "late dye leakage" had a late leakage size more than three times larger than basic leakage size but did not satisfy the criteria of "early dye leakage". MA with "no leakage" had neither early nor late dye leakage. The retinal thickness of the region of each MA was measured with embedded calipers in the OCT B-scan image, and the results were compared between groups. We also compared the sensitivities of the three aforementioned types of MAs in CFP, OCTA, and MC images using the aforementioned methods.

Comparison of MAs in the visibility of dye leakage in OCTA and MC-SLO images. To investigate
whether the degree of fluorescence leakage from MAs affects the detection of MAs in OCTA and MC-SLO, the detection rates of MAs with early and late dye leakage in each image were compared. Two examiners (T.S. and H.T.) detected MAs in OCTA (6 × 6 mm) and MC-SLO images in cases of MAs with early and late dye leakage. The sensitivity (detection rate) of both types of MAs was calculated, and the results were compared between OCTA and MC-SLO.
Comparison of color tone between MA and retinal hemorrhage on MC-SLO images. Because many MAs appear as green dots and retinal hemorrhages appear as red dots on MC-SLO images, the ratio of green and red tones between MAs and retinal hemorrhage was analyzed. From eight eyes containing four or more MAs and retinal hemorrhage, four of each finding were selected on the basis of size order, resulting in a total of 32 spots, and then extracted and analyzed. The color tone of the findings was analyzed using the Wayne Resband mode in ImageJ software (NIH, Bethesda, MD, USA) to obtain the values of red, green, and blue tones. The green/red ratios were calculated and compared.