Microvascular Properties of Dissociated Optic Nerve Fiber Layer (DONFL) After Internal Limiting Membrane Peeling

Abstract

Purpose

To investigate the microvascular properties of dissociated optic nerve fiber layer (DONFL) after internal limiting membrane (ILM)-peeling in the treatment of epiretinal membrane (ERM) using optical coherence tomography angiography (OCTA).

Methods

A prospective, bicentric study of 52 eyes undergoing pars plana vitrectomy (PPV) with ERM- and ILM-peeling for symptomatic ERM. Patients were visited before surgery (baseline), as well as four weeks (V1) and 12 weeks (V2) after surgery. A peer-review-approach of four independent graders and a semi-automated computer-assisted algorithm were conducted to define areas as DONFL, peri-DONFL and non-DONFL. In all areas, perfusion density, vessel density and vessel distance of the superficial and deep capillary plexus were assessed using OCTA.

Results

Of the 52 eyes, a total of 48 eyes reached postoperative visits and were included in further analyses. The incidence of DONFL was 39.58% at V1 and 52.08% at V2. Both in the superficial and the deep vascular plexus, perfusion density and vessel density of DONFL areas were lower than in peri-DONFL and non-DONFL areas, while vessel distance was higher (P < 0.05). Moreover, in the deep capillary plexus vessel density at V1 and V2 and perfusion density at V1 were higher and vessel distance at V1 was lower in peri-DONFL than in non-DONFL areas (P < 0.05).

Conclusions

DONFL is accompanied by a shift within the microvascular network, with capillaries presumably being displaced from DONFL into peri-DONFL areas. These findings support the thesis of a more extensive reorganization of retinal structures that goes markedly beyond the retinal nerve fibre layer.

"Dissociated optic nerve fiber layer” (DONFL), sometimes also referred to as “dimples”, was first described by Tadayoni et al.¹ and refers to the presence of dark, arc-shaped streaks at the posterior pole that follow the course of the retinal nerve fiber layer (RNFL). The original description of the phenomenon was made using blue filter fundus photography, but to a lesser extent it can also be visualized by clinical examination or with color fundus photography.² By now, there is evidence and broad consensus that DONFL is a consequence of internal limiting membrane (ILM) peeling,³ a surgical step often performed as part of pars plana vitrectomy in vitreoretinal interface conditions such as epiretinal membrane (ERM) and macular hole. The incidence of DONFL after ILM peeling detected on fundus photography described in the literature varies between 10.1% and 62.2%.^2–4

The ILM consists of extracellular matrix components, as typically found in basement membranes, as well as the foot plates of the retinal Müller cells.⁵ Consequently, during ILM peeling, the foot plates of the Müller cells are also removed, which is why alterations to the Müller cells and their metabolic, homeostatic and structural functions for the retina can be assumed. The structural changes caused by ILM peeling, which also affect the development of DONFL, are the subject of current research. Although the name suggests that this is merely a dissociation at the level of nerve fibers, recent studies suggest that the alterations within the retina may be deeper than previously thought.^6–8 Moreover, it has not yet been conclusively clarified whether the development of DONFL is also associated with a functional disadvantage for the patient. Most studies conducted on this topic have concluded that apart from the morphological changes, there are no functional differences between patients developing DONFL and those who don't.^3,9,10 Few studies claim a functional disadvantage caused by DONFL-like changes.⁶ However, the comparability of postoperative functional development between eyes with and without DONFL may be distorted. Nukada et al.⁶ found that even eyes without typical DONFL appearance seen in color fundus photography can show similar changes in OCT, hence a potentially negative functional effect of the peeling could also affect eyes without DONFL.

Further studies are needed to better understand the exact pathophysiology of DONFL and to better assess the exact magnitude of the condition. In the following study, we therefore investigated the extent to which DONFL is accompanied by changes of the microvascular properties of the retina using OCT angiography.

Methods

This is a prospective study performed at two tertiary clinics. The study was approved by both local ethics committees (the University Medical Center Göttingen, Ethics Committee of the Westphalia-Lippe Medical Association and the University of Münster) and all patients gave written consent. The data processing and research was carried out in conformity with the tenets of the Declaration of Helsinki and its amendments.

Study Population

A total of 55 patients were consecutively enrolled between November 2020 and September 2022. For study inclusion, a symptomatic ERM in which vitrectomy with ILM-Peeling was indicated, was required. Exclusion criteria comprised myopia or hyperopia of more than ±6 diopters, as well as coexistence of other macular pathologies such as macular hole or diabetic macular edema. If a patient missed one of the planned postoperative appointments, their data was used in further analyses according to the principle last observation carried forward. If both visits were missed, data had to be excluded from the study.

Study Protocol

The first appointment in this study is the date on which the indication for surgical intervention was made (baseline). A multimodal approach ensured the diagnosis of ERM and any possible concomitant diseases were ruled out. After performing pars plana vitrectomy, repeat visits four weeks (V1) and 12 weeks (V2) later were carried out. A series of examinations were performed at all visits. These included determining the best-corrected visual acuity (BCVA), which was measured using the ETDRS chart and converted to logMAR for further statistical analysis. A complete examination of the anterior segment using slit lamp microscopy and the posterior segment using indirect ophthalmoscopy was performed. Multimodal imaging included spectral-domain optical coherence tomography (OCT) of the macula (Spectralis, Heidelberg Engineering GmbH, Heidelberg, Germany) and 12 × 12 mm swept-source OCT angiography (OCTA) with the Plex Elite 9000 (Carl Zeiss Meditec, Inc., Dublin, CA, USA).

Pars plana vitrectomy was performed either with local or intubation anesthesia. Using a 25-gauge approach, core and peripheral vitrectomy were followed by posterior vitreous detachment. After injecting a suitable dye, both the ERM and ILM were peeled with a forceps, which sometimes required multiple applications of the dye.

Data Analysis

DONFL Assessment

The presence of DONFL at baseline, V1 and V2 was evaluated using en-face structural imaging of the superficial vascular plexus by a peer review process. The process comprised an assessment of each image by four independent retina specialists at the Department of Ophthalmology, University Hospital Göttingen. Using the image editing software Paint (Microsoft Corporation, Redmond, WA, USA), each grader manually marked the area in which DONFL was apparent. Using MATLAB (MathWorks, Inc., Natick, MA, USA), all four marked en-face images of each individual OCTA were overlapped. The respective areas were then colorized according to the correspondence between the specialists. Areas were classified as DONFL-positive if at least three of the four graders had independently marked it (Fig. 1A2).

Figure 1.

DONFL assessment on OCTA en-face structure image (A) and preparations to assess perfusion parameters on OCTA en-face angio image (B) of the same eye. Original en-face structure image of the superficial vascular plexus (A1). Same image with colorized overlay showing agreement between the specialists (A2; orange: one grader; yellow: two graders; green: three graders; blue: four graders). Green and blue areas were therefore considered a DONFL-positive target area. Automated detection of DONFL dimples (A3). Final DONFL mask within the inner and outer ring of a fovea centered ETDRS grid (A4; green: DONFL; light gray: peri-DONFL; dark gray: non-DONFL). Original en-face OCTA image of the superficial vascular plexus with ETDRS grid centered at the fovea (B1). Binarized image with each vessel pixel displayed white and each non-vessel pixel displayed black (B2). Skeletonized image with vessel diameter reduced to one pixel (B3). Color-coded distance map showing vessel distance (B4): pixels within vessels are presented in white to red depending on their increasing distance to the vessel border, while pixels not associated with vessels are displayed black to light blue with increasing distance to the vessel border.

Now, an automated approach was used to render out the DONFL dimples as accurately as possible. To achieve this, the en-face structural images with a size of 800 × 800 pixels were first standardized: The gray values were approximated to the gray values in their surrounding area using a standardized formula in order to achieve uniform brightness and reduce noise across the image. First, each pixel's gray value was multiplied by 150 and divided by the median of the gray values of a kernel with a size of 21 × 21 pixels. Next, the average gray value within a 5 × 5 pixel kernel was calculated for each pixel position in order to reduce noise. Within the resulting image, seed points—serving as starting points to mark the DONFL dimples—were then defined as those pixels with a gray value <135. The seed points were then subjected to a growing process in which all pixels which were adjacent to pixels already marked and had a gray value <145, were also marked. This growing loop was repeated until no more pixels were detected. The dimple markings were smoothed using the morphometric operation closing (Fig. 1A3). In the final step, a DONFL mask was generated by transferring the automated preparation of the DONFL dimples as well as the DONFL areas established by grader agreement. The DONFL mask was resized to 1024 × 1024 pixels to ensure consistency in resolution with the en-face angio image. Markings were limited to the corresponding area of the inner and outer ring of an ETDRS grid centered at the fovea centralis within the en-face angio image (Fig. 1B1). Within the DONFL mask (Fig. 1A4), three different areas were therefore defined: DONFL (the automated marked dimples, green), peri-DONFL (areas defined by the graders, in which no dimple was marked, light gray) and non-DONFL (remaining areas, dark gray). Finally, we calculated distribution of the marked DONFL dimples separately for the areas corresponding to the superior, inferior, nasal, and temporal sector of the ETDRS grid.

Measurement of Microvascular Properties

For further analysis of the microvascular properties, 12 × 12 mm en-face OCTA angio images of the superficial vascular plexus and the deep vascular plexus were analyzed using MATLAB (MathWorks, Inc.). To measure perfusion density, a binarization of the image was performed. First, a “vesselness” filter was applied to highlight the vessels.^11,12 Next, binarization classified a pixel as a vessel pixel (white, Fig. 1B2) if either the gray value of the pixel in the original OCTA angio image was >130; or the gray value of the pixel in the “vesselness” image measured >130 and the gray value of the pixel in the original OCTA angio image was greater than the mean value of the surrounding 61 × 61 pixels. Perfusion density was then defined as the proportion of white pixels within the analyzed area. The binarized image allowed to assess the vessel distance. For each pixel, the distance to the nearest vessel border was calculated in eight directions and the shortest distance was recorded¹³ and displayed as a vessel distance map (Fig. 1B4). The mean value of the shortest vessel distance in the analyzed sector accounted for the vessel distance value. The binarized image was skeletonized, reducing each vessel to a diameter of 1 pixel by deleting the pixel with the shortest distance to the vessel edge within each 2 × 2 pixel kernel with four white vessel pixels. The proportion of all white pixels within the skeletonized image then yielded the vessel density. Finally, the three microvascular parameters were calculated for the area corresponding to the inner and outer ring of the EDTRS grid, as well as for the areas corresponding to DONFL, peri-DONFL and non-DONFL within the DONFL mask.

Statistical Analysis

The statistical analysis was carried out with GraphPad Prism 10 (GraphPad Software, Inc., San Diego, USA) and Excel (Microsoft Corporation, Redmond, WA, USA). Descriptive analysis included sex, age, visual acuity progression, incidence of DONFL and distribution of DONFL within the ETDRS grid. Differences between distributions of DONFL within the ETDRS grid and differences of perfusion parameters between visits and eyes as well as between DONFL, peri-DONFL and non-DONFL at V1 and V2, were calculated using Wilcoxon signed rank test and Friedman test. Because the number of predefined OCTA parameters was limited, no formal correction for multiple testing was applied. Results at a significance level of p ≤ 0.05 were considered statistically significant. To test the re-test reliability of DONFL grading, 10% of all 137 OCTA scans (i.e., 14 images) were randomly selected and added twice to the data set to be assessed by the four graders. Fleiss’ Kappa was used to test for interrater reliability, Cohen's Kappa was used to test for intrarater reliability.

Results

Study Population

A total of 52 patients were initially enrolled in this prospective study, 28 (53.9%) being recruited in the Department of Ophthalmology, University Hospital Göttingen and 24 (46.2%) in the Ophthalmology Department at St. Franziskus-Hospital Münster. Twenty-five patients (48.1%) were female, and 27 patients (51.9%) were male, with an average age (mean ± SD) of 71.5 ± 7.7 years. A total of four patients (7.7%) had to be excluded from the study because they discontinued the study after baseline. Another four patients, one of whom (1.9%) suffered from retinal detachment after V1 and three of whom (5.8%) discontinued the study after V1, were included in the study in accordance with the last observation carried forward principle. A total of 48 patients were therefore available for statistical analyses.

Visual Acuity Outcome

Average BCVA (mean ± SD) of the whole cohort at baseline was 0.42 ± 0.22 logMAR and improved to 0.39 ± 0.29 logMAR at V1 and 0.30 ± 0.29 logMAR at V2. Sub-group analysis showed that patients not developing DONFL progressed from 0.41 ± 0.25 logMAR at baseline to 0.36 ± 0.29 logMAR at V1 and 0.35 ± 0.27 logMAR at V2. Those who developed DONFL improved from 0.42 ± 0.18 logMAR to 0.42 ± 0.29 logMAR at V1 and 0.27 ± 0.30 logMAR at V2. However, the functional outcome between eyes with and without DONFL did not show statistical significance (P > 0.05). Furthermore, there was no significant correlation between the size of DONFL area and BCVA outcome or change in BCVA from baseline to V2.

Incidence of DONFL and Distribution Within the ETDRS Grid

At baseline, none of the eyes showed DONFL. At V1, 19 out of 48 patients (39.58%) developed DONFL. At V2, the incidence of DONFL increased to 25 patients (52.08%). At V1 the median of DONFL area of the 25 patients developing DONFL measured 7.48% within the temporal sector of the ETDRS grid, whereas in the nasal (1.21%), superior (2.06%), and inferior sector (0.72%), DONFL was less apparent. At V2, the median of DONFL area increased to 13.52%, 7.46%, 10.29% and 7.61% in the temporal, nasal, superior and inferior sectors, respectively (P < 0.005). Maximal DONFL area at V1 and V2 measured 20.47 and 21.26% in the temporal sector, 22.55 and 25.34% in the nasal sector, 24.02 and 28.99% in the superior sector, and 13.99 and 20.09% in the inferior sector. A more detailed distribution of DONFL area is displayed as boxplots in Figure 2B and in the Table.

Figure 2.

DONFL distribution. Examples of DONFL (A) displayed on original OCTA en-face structure image (upper rows) with DONFL dimples highlighted green (lower rows) at visit V1 after four weeks and visit V2 after 12 weeks after surgery. Boxplots represent distribution of DONFL area across the ETDRS sectors at each visit (B). **** P < 0.0001, *** P < 0.005.

Table.

Distribution and Microvascular Properties of DONFL

	V1 (4 Weeks)			V2 (12 Weeks)
	Median	IQR (Q1 – Q3)	n	Median	IQR (Q1 – Q3)	n	P * Vs. V1
Distribution of DONFL
DONFL area [%]
ETDRS ring	4.12	6.04 (0.61 – 6.64)	25	9.47	7.66 (5.84 – 13.50)	24	<0.0001
Superior	2.06	4.48 (0.00 – 4.48)	25	10.29	10.60 (6.26 – 16.86)	24	<0.0001
Inferior	0.72	6.19 (0.00 – 6.19)	25	7.61	9.96 (3.05 – 13.00)	24	<0.0001
Temporal	7.48	14.20 (0.00 – 14.20)	25	13.52	13.39 (3.63 – 17.01)	24	<0.0001
Nasal	1.21	6.32 (0.00 – 6.32)	25	7.46	12.96 (0.70 – 13.66)	24	0.004
Perfusion of superficial capillary plexus
Perfusion density [%]
DONFL	28.77	5.21 (25.37 – 30.58)	19	27.75	5.82 (23.36 – 29.18)	24
Peri-DONFL	40.98	3.00 (39.30 – 42.30)	19	42.71	4.47 (39.24 – 43.98)	24
Non-DONFL	41.89	5.48 (37.95 – 43.43)	19	41.31	6.42 (37.94 – 44.36)	24
P†		<0.0001			<0.0001
Vessel density [%]
DONFL	19.13	3.27 (16.84 – 20.11)	19	18.30	3.01 (16.21 – 19.22)	24
Peri-DONFL	22.83	2.16 (21.29 – 23.45)	19	22.81	2.84 (21.22 – 24.06)	24
Non-DONFL	22.96	2.75 (21.24 – 23.99)	19	22.68	2.23 (21.75 – 23.98)	24
P†		<0.0001			<0.0001
Vessel distance [pixel]
DONFL	1.66	0.29 (1.57 – 1.86)	19	1.72	0.23 (1.62 – 1.84)	24
Peri-DONFL	1.42	0.11 (1.40 – 1.50)	19	1.43	0.17 (1.37 – 1.54)	24
Non-DONFL	1.42	0.17 (1.38 – 1.55)	19	1.45	0.15 (1.37 – 1.52)	24
P†		<0.0001			<0.0001
Perfusion of deep capillary plexus
Perfusion density [%]
DONFL	32.83	4.93 (30.71 – 35.64)	19	34.19	4.79 (32.03 – 36.82)	24
Peri-DONFL	37.73	4.20 (35.10 – 39.30)	19	37.96	3.97 (35.22 – 39.19)	24
Non-DONFL	36.15	4.69 (34.21 – 38.90)	19	37.18	3.17 (35.26 – 38.43)	24
P†		<0.0001			<0.0001
Vessel density [%]
DONFL	22.50	2.86 (21.77 – 24.63)	19	23.57	2.48 (22.11 – 24.59)	24
Peri-DONFL	24.39	1.75 (23.86 – 25.61)	19	24.95	1.68 (23.85 – 25.53)	24
Non-DONFL	23.95	1.91 (22.80 – 24.71)	19	24.37	1.52 (23.34 – 24.86)	24
P†		<0.0001			<0.0001
Vessel distance [pixel]
DONFL	1.45	0.16 (1.34 – 1.50)	19	1.40	0.13 (1.33 – 1.47)	24
Peri-DONFL	1.34	0.09 (1.30 – 1.39)	19	1.32	0.09 (1.29 – 1.38)	24
Non-DONFL	1.37	0.11 (1.32 – 1.43)	19	1.35	0.09 (1.31 – 1.39)	24
P†		<0.0001			<0.0001

Q1, 25^th percentile; Q3, 75^th percentile.

^*Wilcoxon signed rank test.

^†Friedman test.

Two tests were conducted to calculate interrater reliability. With regard to the detection of DONFL, Fleiss’ kappa was κ = 0.80, implying a substantial agreement according to Landis and Koch (1977).¹⁴ In terms of the agreement between the DONFL areas marked in each case, Fleiss’ kappa was κ = 0.59, which corresponds to moderate agreement. Intrarater reliability was calculated for each grader, regarding the detection of DONFL. Cohen's kappa was κ = 0.81, κ = 0.84, κ = 1.0, and κ = 0.84 for grader 1, 2, 3, and 4, respectively, resulting in an almost-perfect agreement according to Landis and Koch.¹⁴

Microvascular Differences Within the Entire EDTRS Grid at Baseline, V1 and V2

First, the perfusion parameters within the entire ETDRS grid were examined not yet separated into areas with and without DONFL. Differences between the study eye and the healthy fellow eye were tested, as well as changes from baseline to V1 and V2.

Regarding the superficial vascular plexus, only a minor difference was observed between the study eye and the fellow eye, as vessel distance was higher in the study eyes than the fellow eyes at baseline (P = 0.0459). This finding was not detectable anymore after surgery.

However, in terms of the deep vascular plexus, a different observation was made. Although the perfusion density did not show any differences, the vessel density was significantly lower in the study eyes compared to the fellow eyes at baseline (P = 0.0006). Moreover, study eyes showed an increasing vessel density over time. While the median of vessel density was 22.83% at baseline, it rose to 23.68% at V1 (P = 0.0082) and 23.86% at V2 (P = 0.0003), respectively. A similar picture was observed regarding the vessel distance. At baseline, vessel distance was greater in study eyes compared to the fellow eyes (P = 0.0006). Furthermore, the vessel distance decreased in the study eyes from 1.45 pixels at baseline to 1.39 pixels at V1 (P = 0.0073) and 1.37 pixels at V2 (P = 0.0002). These findings are illustrated in Figure 3 and Supplemental Table S1.

Figure 3.

Microvascular differences within the entire EDTRS grid. Boxplots display differences between study eyes and fellow eyes at baseline, V1 and V2 in terms of perfusion density (A1 and B1), vessel density (A2 and B2), and vessel distance (A3 and B3) of the superficial vascular plexus (top row, A) and the deep vascular plexus (bottom row, B). *P < 0.05; **P < 0.01; ***P < 0.001.

Microvascular Differences Between DONFL, Peri-DONFL, and Non-DONFL Areas at V1 and V2

Differences in perfusion density, vessel density, and vessel distance in the superficial, as well as the deep vascular plexus between DONFL area, peri-DONFL area, and non-DONFL, areas were tested at visits V1 and V2. With regard to the superficial vascular plexus, there was a significant difference of all parameters when DONFL areas were compared to peri-DONFL and non-DONFL areas. At V1, the perfusion density (median) in DONFL areas was 28.77% and in peri-DONFL and non-DONFL areas 40.98% and 41.89%, respectively (P < 0.0001). A similar picture was seen in V2, where perfusion density amounted to 27.75% in DONFL areas, 42.71% in peri-DONFL and 41.31% in non-DONFL areas (P < 0.0001). A similar observation was made in terms of the deep vascular plexus. At V1, perfusion density was 32.83% in DONFL areas, whereas peri-DONFL and non-DONFL accounted for 37.73% and 36.15%, respectively (P < 0.0001). This was also seen at V2, where perfusion density in DONFL areas was significantly lower than in peri-DONFL and non-DONFL areas, being 34.19%, 37.96%, and 37.18%, respectively (P < 0.0001). Interestingly, at V1, the perfusion density of the deep vascular plexus was also significantly higher in peri-DONFL compared to non-DONFL (P = 0.0181).

Vessel density (median) was distributed in a similar manner in the superficial vascular plexus. Here too, at V1 the value in DONFL areas was significantly lower than in the peri-DONFL and non-DONFL areas, amounting to 19.13%, 22.83% and 22.96% (P < 0.0001). Also, in V2 vessel density was lower in DONFL areas than in peri-DONFL and non-DONFL areas, being 18.30%, 22.81% and 22.68%, respectively (P < 0.0001). In the deep vascular plexus, at V1 vessel density was again lower in DONFL areas (22.50%) than in peri-DONFL (24.39%, P = 0.0002) and non-DONFL areas (23.95%, P = 0.0108). At V2, vessel density was 23.57% in DONFL areas and therefore lower than in peri-DONFL (24.95%, P < 0.0001) and non-DONFL areas (24.37%, P = 0.0006). Interestingly, both at V1 and V2 vessel density of the deep vascular plexus was significantly higher in peri-DONFL than in non-DONFL (P = 0.0001 and P = 0.0269, respectively),

Vessel distance (median) was distributed in a similar manner in the superficial vascular plexus, but with an inverse relationship between the areas. At V1, vessel distance was significantly higher in DONFL areas than in peri-DONFL and non-DONFL areas, amounting to 1.66 pixels, 1.42 pixels and 1.42 pixels, respectively (P < 0.0001). At V2, vessel distance was again higher in DONFL areas than in peri-DONFL and non-DONFL areas, being 1.72 pixels, 1.43 pixels, and 1.45 pixels, respectively (P < 0.0001). Regarding the deep vascular plexus, at V1 the vessel distance was again higher in DONFL areas (1.45 pixels) than in peri-DONFL (1.34 pixels, P < 0.0001) and non-DONFL areas (1.37 pixels, P = 0.0071). Once again, it became apparent here too that vessel distance was significantly lower in peri-DONFL than in non-DONFL (P = 0.0003). At V2, DONFL areas had greater vessel distance (1.40 pixels) than peri-DONFL (1.32 pixels, P < 0.0001) and non-DONFL areas (1.35 pixels, P = 0.0005). A graphical representation of these analyses is depicted in Figure 4 and an overview of perfusion data with median and IQR can be found in the Table.

Figure 4.

Perfusion within DONFL, peri-DONFL and non-DONFL areas. Boxplots display differences between DONFL (green), peri-DONFL (gray), and non-DONFL (dark gray) areas at V1 and V2 in terms of perfusion density (A1 and B1), vessel density (A2 and B2), and vessel distance (A3 and B3) of the superficial vascular plexus (top row, A) and the deep vascular plexus (bottom row, B). *P < 0.05; **P < 0.01; ***P < 0.001; **** P < 0.0001.

Discussion

The aim of this study was to investigate the pathophysiology and extent of DONFL after vitrectomy with ILM peeling, which remain not fully understood, with a particular focus on examining the microvascular properties of the retina.

Incidence and Functional Relevance

First, we confirmed that correctly indicated surgical treatment of ERM is associated with a postoperative increase in visual acuity, which generally reinforces the relevance of this surgical procedure. In accordance with most previous studies, the development of DONFL also appeared to remain without functional impairment in our cohort, as there was no significant difference in visual acuity between DONFL-positive and DONFL-negative patients.

The incidence of DONFL described in this study, at 39.58% after four weeks and 52.08% after 12 weeks, lies within the middle range of incidences described before. However, current studies investigating DONFL are exploratory in nature, and it is difficult to compare incidences because of the wide variety of definitions and techniques used in the studies published. Previous studies using fundus photography to assess DONFL reported incidences ranging from 10.1% to 62.2%.^2–4 While in initial studies DONFL was mainly assessed by fundus photography, other studies show that the incidence of DONFL-like changes within the RNFL may be even higher when looking at other imaging modalities such as OCT scans. Kim et al.¹⁵ found DONFL in OCT in 80.3% of the cases, Steel et al.¹⁶ described DONFL-like changes in OCT in even 100% of 39 ILM-peeled eyes, and Nukada et al.⁶ reported DONFL-like changes in OCT in 87.5% of those eyes without DONFL on colour fundus photography. Because the assessment of DONFL in the present study was performed using OCTA en face structure images and was not compared to findings in OCT, it is possible that the incidence of DONFL-like changes in OCT could have been higher in this study as well. Also, the fact that previous studies claimed that similar changes can be seen in OCT even in eyes without typical DONFL appearance on fundus photography makes it difficult to examine for functional disadvantages caused by DONFL, because presumably healthy eyes are used as a control group even though DONFL-like changes may also be apparent in OCT. Furthermore, other studies confirm a certain increase in DONFL incidence over a period of up to six months after surgery, meaning that the incidence described also depends on the time of the data collection. However, most of the eyes developing DONFL seem to do so up to three months after surgery.¹⁵ In addition, the treated retinal pathology may also have an influence on the incidence of DONFL. Kim et al.¹⁷ found an incidence of DONFL of 75.0% after treatment of a macular hole and 22.0% after treatment of an ERM. For these reasons, it is difficult to compare the incidence of DONFL determined in this study with incidences from previous studies. Nevertheless, it can be said that this finding occurs frequently, which is why a more precise understanding is important.

Previous studies have already shown that the distribution of DONFL is concentrated in particular in the temporal portion of the macula and increases over time throughout all sectors.^18,19 In the present study, we were able to confirm this observation. At the first postoperative visit after four weeks, the distribution of DONFL was most pronounced in the temporal sector and increased in all sectors after 12 weeks. One possible explanation for this observation is that ILM peeling is presumed to cause a shift of the macula toward the optic disc. The temporal nerve fibres appear to be more susceptible to mechanical stress than the nasal fibres, whose proximity to the optic disc may have a stabilizing effect.^20,21 The restructuring of retinal morphology as a result of ILM peeling is also the subject of discussion in the context of the development of DONFL. Removing the foot processes of Müller cells could potentially damage the integrity of the retina. However, it is likely that the development of DONFL does not result in actual loss of further retinal structures. Rather, ILM peeling is presumably followed by a separation of RNFL bundles and potentially other retinal layers.^2,3 A few studies have given indication that the alterations affect not only the RNFL, but also deeper structures of the retina up to the inner plexiform layer.^6–8

Microvascular Alterations

Regardless of the presence of DONFL, the microvascular analyses in the present study showed that the reduced perfusion observed preoperatively in comparison to the fellow eye became similar to that of the fellow eye after ERM peeling. We observed no difference within the superficial vascular plexus, however, the deep vascular plexus showed an increase of vessel density and a decrease of vessel distance in eyes treated for ERM. Microvascular changes following ERM or ILM peeling remain controversial. In contrast to our results, Mastropasqua et al.²² reported a reduction of vessel density in the superficial capillary plexus. In line with the present study, Lauermann et al.²³ previously measured a reorganization of the deep vascular capillary plexus with increasing perfusion after ILM peeling in eyes with macular holes.

Retinal vascular perfusion in the context of DONFL has not yet been studied in detail. Only Kim et al.¹⁷ investigated changes of the perifoveal vessel density in relation to the development of DONFL. In their study, vessel density in the temporal sector, which is typically the most common location for DONFL, decreased significantly after ILM peeling.¹⁷ Although the study provided some evidence of microvascular changes after ILM peeling, the role of DONFL in relation to the retinal microperfusion remained unclear. To this date, no such studies have more precisely investigated the microvascular alterations that may occur in the case of DONFL.

By examining the eyes that developed DONFL and distinguishing more precisely between DONFL areas and surrounding retinal areas not affected by DONFL, new insights into the microvascular perfusion of the retina were gained in the present study. Both in the superficial and deep vascular plexus, a reduced flow was observed in the DONFL areas, measured as decreased perfusion density and vessel density, as well as higher vessel distance, compared to the peri-DONFL and non-DONFL areas. This difference remained relatively stable when comparing visits four weeks and 12 weeks after surgery.

A closer examination of the retinal microperfusion showed that DONFL development also caused a shift in the vascular network in the deep capillary plexus, which is anatomically located between the inner nuclear layer and the outer plexiform layer. While perfusion within the DONFL areas was reduced, areas immediately adjacent to the DONFL areas showed increased perfusion compared to the rest of the surrounding retina. This observation is consistent with the finding described in this study that ILM peeling does not lead to a deterioration in perfusion when considering the entire EDTRS grid (i.e., without distinguishing between DONFL and non-DONFL areas). This study thus confirms the above-mentioned thesis that the structural changes associated with the development of DONFL are not limited to the retinal nerve fiber layer, but seem to extend far into the deeper layers of the retina. This also means that the term “DONFL” is not entirely correct, as it suggests that the change only occurs at the level of the nerve fibers.

DONFL may therefore reflect an underlying structural and vascular remodeling of the retina rather than simple surface irregularities. This finding is also clinically relevant, as it underlines the value of DONFL as a biomarker for the comparison of different surgical techniques and peeling strategies: a lower incidence of DONFL could therefore be interpreted as a surrogate marker for fewer structural changes in the retina when comparing different surgical approaches. Ehrhardt et al.,²⁴ for example, investigated the extent to which the conventional ILM peeling technique influences the incidence of DONFL compared to a newer temporal inverted ILM flap technique, whereby the latter technique led to a reduction of DONFL in the nasal area. Future studies could similarly investigate the extent to which different approaches can reduce the incidence of DONFL and thus structural retinal changes. Our findings also underline the significance of DONFL with regard to the postoperative outcome for patients. Even though most studies show that the occurrence of DONFL does not lead to functional losses,^3,9,10 individual studies come to a different conclusion. Nukada et al.⁶ differentiated between microperimetry test points in terms of their position to DONFL areas. They found that the sensitivity was slightly reduced in those DONFL areas with deeper structural deficits on OCT compared to areas without DONFL.⁶ Their study underlines the suggestion that it is probably necessary to take a more differentiated approach to evaluating functional changes caused by DONFL. It may not be sufficient to consider only visual acuity and global microperimetry results, but rather a more precise differentiation between affected and unaffected areas within the macula should be assessed. Furthermore, it is probably not sufficient to compare eyes with typical DONFL, as described in fundus photographs, with eyes without DONFL after ILM peeling. Studies have shown that even eyes without DONFL in fundus photographs often show typical changes in OCT,⁶ meaning that these eyes should probably not be used as a control group. This also demonstrates that a more differentiated approach is necessary.

The role of microvascular rearrangements seen in our study could be of particular significance in eyes already suffering from concomitant diseases and therefore having a lower reserve to microvascular und structural changes. Diseases such as glaucoma or diabetic retinopathy are systematically excluded in most DONFL studies and are therefore less well investigated in terms of their occurrence and functional outcome. Sakimoto et al.²⁵ examined highly myopic eyes after macular surgery with ILM peeling and found DONFL dimples in all cases in the OCT en-face, demonstrating that comorbidities could play a decisive role. Future studies are necessary to investigate other comorbidities and their influence on incidences and functional outcomes.

There are a few limitations to this study. First, only patients who were treated for ERM were included. However, ILM peeling is also used for other conditions, such as the treatment of macular holes, vitreoretinal traction, or retinal detachment. As a result, it remains unclear whether the findings of this study can be readily applied to these diseases. Moreover, there's the surgical method itself: even though it's a pretty standardized procedure, there's a certain amount of variability because of the bicentric study design and the fact that the procedures were done by various surgeons. In addition, the grading of the DONFL area was subject to a certain degree of subjectivity. This circumstance was counteracted by the peer review process involving a total of four graders, whereby only those areas with agreement from at least three graders were classified as DONFL areas, as well as the automated detection of the DONFL dimples themselves. Interrater and intrarater reliability showed moderate to almost perfect agreements.

Conclusions

DONFL is a phenomenon that often occurs as a result of ILM peeling. The exact pathophysiology and, in particular, the extent of this finding are not yet fully understood. Recent studies suggest that the structural changes associated with the development of DONFL extend markedly beyond the level of the nerve fiber layer. A more detailed examination of retinal microperfusion, conducted for the first time in this study, may support this hypothesis. DONFL appears to be associated with a shift of the vascular network from the DONFL areas to the surrounding tissue, which also affects the deep vascular plexus located at the level of the inner nuclear layer and the outer plexiform layer.