Evaluation of the inner retinal layers following traditional and modified ILM peeling surgery using automatic CMDS quantification algorithm

Idiopathic macular hole (IMH) is a vitreoretinal disease that causes severe visual impairment, and the pars plana vitrectomy (PPV) combined with internal limiting membrane peeling (ILMP) is the traditional treatment for IMH. Since the dissociated optical nerve fiber layer (DONFL) is observed in the eyes after the surgery, a modified surgery is put forward to minimize the injury. However, the extent to which the modified surgery reduced damage is not clear. A novel automated image processing system was developed in this study to quantify the damage after the surgery, revealing how the modified surgery reduced the DONFL. Optical coherence tomography (OCT) images and visual acuity were collected from 65 eyes that underwent a vitrectomy with traditional and modified ILM peeling procedure at Zhejiang Provincial People’s Hospital preoperatively. We found that DONFL occurred in all 65 eyes of both the traditional and modified ILM peeling groups. The concentric macular dark spots (CMDS) were automatically measured by the image processing system. The mean total area of CMDS was smaller in the modified ILM peeling group (0.224 ± 0.110 mm2) than that of the traditional ILM peeling group (0.385 ± 0.237 mm2), and there is a significant difference between the two groups (p < .05). The morphological difference in the temporal quadrant of CMDS between the two surgery technology groups was significant (p < .001). This study will provide a clinical research basis for the improvement of macular hole surgery.


INTRODUCTION
Idiopathic macular hole (IMH) is a vitreoretinal disease that causes severe visual impairment.The overall incidence is approximately 4 to 8.7 cases per 100,000 people per year, which would be much higher among women and the elderly. 1,2With the progression of IMH, the best-corrected visual acuity (BCVA) would typically decrease to 20/200 (Snellen), causing serious vision impairment. 3Pars plana vitrectomy (PPV) combined with internal limiting membrane peeling (ILMP) and intravitreal gas tamponade is the standard treatment for IMH, with a high anatomical closure rate of 80-100%. 4,5And ILMP is one of the key steps, which can remove the tangential traction exerted by ILM. 6 However, as the basal membrane of Müller cells, ILM plays a role in homeostasis and maintenance of inner retinal architecture. 7ILMP may cause various complications, such as swelling of the arcuate retinal nerve fibre layer, dissociated optical nerve fibre layer (DONFL), macular retinal displacement, etc. 8,9 The DONFL is observed as numerous arcuate striae along the nerve fibre direction within the posterior pole, slightly darker than the surrounding retina on blue-filter photographs, first defined by Tadayoni in ERM patients after ILMP. 10 With the development of multimodal imaging technology, colour fundus photography, fundus autofluorescence, red-free fundus photography, scanning laser ophthalmoscope, and spectral-domain optical coherence tomography B-scan all can be used to observe DONFL. 11SD-OCT Cscan (en face OCT) offers a coronal view of the different layers of the retina, is more distinct and accurate than other images in detecting these retinal changes layer-bylayer. 12The DONFL can be found almost 100% in IMH after ILMP. 8,13or the distribution pattern of DONFL, some literature reported, DONFL mainly occurred on the temporal side of the macula on en face OCT images, 14,15 Our previous study confirmed it. 16The influence of DONFL on the visual function is controversial.Some scholars found it was related to a reduction in central retinal sensitivity and paracentral scotomas.Different surgical techniques were tried, such as peeled ILM reposition, pinch-peel technique, to decrease the incidence of DONFL. 17,18However, DONFL cannot be avoided.Since DONFL presented more in the temporal quadrant.We aimed to explore a new technique to reduce the temporal DONFL.Here, we introduce a modified ILM peeling technique in IMH-"eccentric ILMP," in which the ILM in the temporal quadrant of the macula is peeled less than the nasal side.
Fewer studies count DONFL in each of the central 13 slices in OCT B-scan, or describe it as "mild," "moderate," or "severe" in en-face OCT.All of these assessments had limitations.In our previous study, we developed a MATLAB-based automated system to quantitatively assess the DONFL. 16A novel system using the local adaptive thresholding (LAT) algorithm was developed for the detection and location of the DONFL on en-face OCT images.The system could automatically calculate the area of the concentric macular dark spots (CMDS) and calculate its percentage in each subregion after excluding the pixels on the macular hole, and an analysis report would be then exported.
In this study, we aimed to quantitatively assess DONFL in IMH after modified or traditional ILMP with the use of en face OCT in a novel automated systems, to evaluate the effectiveness of the novel technique in decreasing DONFL.

Ethics and information governance
This research received approval from the Ethics Committee of Zhejiang Provincial People's Hospital.This study follows the principles of the Helsinki Declaration.The OCT scans of all patients were collected centrally by the fundus eye disease center of Zhejiang Provincial People's Hospital.Informed consent to participate in the study was obtained from all of the participants.

Participants and surgery
This retrospective study reviewed the medical records of patients who underwent PPV with ILM peeling for IMHs with a minimum follow-up of 6 months at the Zhejiang Provincial People's Hospital between January 2017 and September 2020.Inclusion criteria included postoperative type 1 closed IMHs with a single surgery.Type 1 closure is defined as MHs closed with some reconstitution of the intralayer structure of the retina. 19Exclusion criteria included other severe macular diseases, myopia less than −6 diopters, any co-morbidities including glaucoma, uveitis, and diabetic retinopathy, history of previous vitrectomy, and poor-quality OCT images during follow-up.
All surgeries were performed by a single surgeon using a 23-gauge transconjunctival vitrectomy system with an Accurus 800 CS Surgical System (Alcon Laboratories, Inc.).The surgical procedure was a standard three-port PPV with triamcinolone acetonide assisted posterior vitreous detachment, followed by removing as much peripheral vitreous as possible.It was first stained with 0.02 mL of indocyanine green (0.025 mg/mL).The ILM in an area of four disc diameters around the fovea would be removed using traditional ILMP technology (Figure 1D), while less ILM in the temporal quadrant would be peeled using mod-ified technology (Figure 1J).After a gas-liquid exchange in the vitreous cavity, sterile air was filled.When needed, concomitant phacoemulsification with an intraocular lens was performed prior to PPV.All patients were asked to maintain a facedown position for a minimum of seven days.The ocular examination included pre-and postoperative BCVA in Snellen visual acuity chart, anterior segment, and fundus examination using a slit lamp and indirect ophthalmoscopy.

OCT scanning
For each eye, a volume scan containing multiple highdensity cross-sectional OCT scans was obtained using an SD-OCT (RTVue, Optovue) with a scanning speed of 70,000 A-scans/s offering an optical axial resolution of 5 μm.A total of 400 unaveraged cross-sectional scans per acquisition, covering a 6 × 6 mm area of the macula, were taken.Only well-cantered images with a signal strength index higher than 6 (out of 10) and without motion or blinking artifacts were analyzed.Three-dimensional reconstructions of en-face OCT images of superficial and deep retinal macular areas were obtained using the 3D Display software (RTVue) (Figure 2).

Automated image processing and qualification
First, an estimation of CMDS distribution characteristics was generated based on the en-face OCT image.The image was divided into four different quadrants, temporal, nasal, superior, and inferior, with the fovea as the center.The radius (r) of CMDS in each quadrant was defined as the smallest radius centered on the fovea that covers all CMDS in a single quadrant.The area of CMDS was defined as the total area of the dark spots on the en-face image."The nerve fiber layer dissociation index" (NFLDI) 16 was used to estimate the intensity of CMDS, which is defined as the ratio of the area of CMDS to the corresponding retinal area (S = 1/4πr 2 ).We developed a Python, OpenCV, 20 and Numpy 21 based program to automatically and quantitatively evaluate the area and NFLDI of concentric macular dark spot (CMDS), which consists of five subfunctions including quadrant determination, image preprocessing, anchor localization and region division, and CMDS detection.For the quadrant determination, the N, S, T, I between the selected image and the template would be calculated, and this variance would help the algorithm automatically determine the direction of the temporal, nasal, superior, and inferior quadrants.For the image preprocessing, the region of interest (ROI) sized 866 × 866 that contained all the CMDS and the macular hole was cropped, and the Gaussian filtering is performed on the ROI by the convolution kernel sized 3 × 3.For the anchor localization and region division, the first step was to convert an RGB color image to a gray image, and then a global threshold segmentation algorithm was used on the gray image.After thresholding segmentation, the gray image is converted to a binary image, and we performed a connected domain algorithm to search for the maximum connected area on the binary image.The maximum connected area represented the macular hole since the center of the macular was the biggest CMDS due to its depth.After locating the area of the macular hole, we divided the ROI into several subregions and marked the centroid of the macular hole as the center point.According to ETDRS, we used two straight lines passing through the macular centroid at angles of 45 • and 135 • to divide into four quadrants: temporal, nasal, superior, and inferior.For CMDS detection, the vitally important step was using the LAT algorithm, which could change the threshold dynamically to segment the image. 22Global thresholding used a fixed threshold for all pixels in the image and only worked when the whole image had a uniform brightness, while LAT is used as an individual threshold for each pixel based on the range of intensity values in its local neighborhood.Hence, LAT could separate desirable foreground pixels from the background and then effectively extract the CMDS on the ROIs. 23Finally, the algorithm could calculate the area of CMDS and calculate its percentage in each subregion after excluding the pixels on the macular hole, and an analysis report would be then exported.
Images were read and analyzed by two libraries, OpenCV and Numpy. 20The program was designed to read all the images under a selected folder, record the names of the images in memory, and process them one by one.The CMDS in different quadrants would be marked as different color.Image information of the CMDS area and its percentage in different quadrants was output in the CSV format text file, each image corresponded to a certain text file.

Statistical analyses
Statistical analyses were conducted with statistics software (IBM SPSS, version 26; IBM Corp.).Categorical and continuous data were expressed in percentage and mean ± standard deviation, respectively.The normality of the data distribution was confirmed using the Shapiro-Wilk test.A p-value of less than .05was statistically significant.

Patient characteristics
This study included 65 eyes of 65 patients (15 men and 50 women, with a mean age of 59.39 ± 9.56 years).Of them, 47 eyes underwent the traditional ILM peeling surgery, and 18 eyes underwent the modified ILM peeling surgery.The mean follow-up period was 6 months.CMDS was found in 65 eyes in both traditional and modified ILM peeling  groups.The detail of characteristics was shown in Table 1 and Figure 3.

Compare the CMDS outcomes of the two surgical methods
The traditional ILM peeling resulted in a larger involvement range of CMDS compared to modified ILM peeling in this study (Figure 4C-F).The mean radius of CMDS was 2.052 ± 0.458 mm in the traditional ILM peeling group,

Characteristics Traditional Modified p Value
The radius of involvement (mm) while the mean radius was 1.721 ± 0.235 mm in the modified ILM peeling group (Table 2).There was a statistically significant difference between the two groups (p < .01).
The mean total area of CMDS was 0.385 ± 0.237 mm 2 in the traditional ILM peeling group, and was 0.224 ± 0.110 mm 2 in the modified ILM peeling group.There is a significant difference between the two groups (p < .01).The NFLDI was 8.064 ± 3.207% in the traditional ILM peeling group, and was 7.094 ± 3.062% in the modified ILM peeling group.There is no significant difference between the two groups (p > .05).

The different distribution patterns of the CMDS after two different surgeries
The radius of CMDS involvement showed significant differences among the four quadrants (Table 3 and Figure 5A).The radius in the temporal quadrant was the largest among the four quadrants in the traditional ILMP group (p < .001)but the radius in the nasal quadrant was the largest among the four quadrants in the modified ILMP group (p < .001).As for the comparison between the same quadrant, there was a statistically significant difference in the radius of CMDS between the traditional ILMP group (2.509 ± 0.533 mm) and the modified ILMP group (1.438 ± 0.274 mm) in the temporal quadrant (p < .001),while no significant difference in the radius of CMDS was found in the other three quadrants (p > .05).The distributed heat map of the CMDS radius was shown in Figure 6.
The area in the modified ILMP group was significantly smaller than the area in the traditional ILMP group (p < .001)(Table 3 and Figure 5B).Moreover, the area in the temporal quadrant was the largest among the four quadrants in the traditional ILMP group (p < .001)but the area in the nasal quadrant was the largest among the four quadrants in the modified ILMP group (p = .042).As for the comparison between the same quadrant, there was also a statistically significant difference in the area of CMDS between the traditional ILMP group (0.801 ± 0.510 mm) and the modified ILMP group (0.235 ± 0.153 mm) in the temporal quadrant (p < .001).The area of CMDS was significantly fewer in the modified ILMP group than in the traditional ILMP group in the inferior quadrant (p = .002).But no significant difference in the area of CMDS was found in the other two quadrants (p > .05).The distributed heat map of the CMDS area was shown in Figure 6.The NFLDI also showed significant differences among the four quadrants (Table 3 and Figure 5C).The NFLDI in the temporal quadrant was the largest among the four quadrants in the traditional ILMP group (p < .001)but the NFLDI in the nasal quadrant was the largest among the four quadrants in the modified ILMP group (p = .002).When compared to the same quadrant, there was no statistically significant difference between the traditional ILMP group and the modified ILMP group in the temporal and nasal quadrants (p > .05),but a significant difference was found in superior and inferior quadrants (p < .05).Further analysis for the distribution of the CMDS in the four quadrants was performed and there was a significant difference between the four quadrants both in the traditional ILMP group and modified ILMP group (p < .001).The distributed heat map of NFLDI was shown in Figure 6.

DISCUSSIONS
In this study, we compared the difference in the morphology and distribution of DONFL caused by conventional ILMP and modified ILMP 6 months postoperatively.A novel system using the LAT algorithm was developed for the detection and location of the DONFL on en-face OCT images.The system could automatically calculate the area of CMDS and calculate its percentage in each subregion after excluding the pixels on the macular hole, and an analysis report would be then exported.The results showed that compared with the traditional ILMP, the mean radius of involvement and area of DONFL caused by the modified ILMP would be smaller.We speculate that the smaller radius and area of DONFL involvement were related to less peeled ILM in the temporal quadrant in the modified ILMP group compared to the traditional ILMP group.We also compared the morphological differences of DONFL between the two groups with different surgery technology in four quadrants: temporal, nasal, superior, and inferior.The results showed that the modified ILMP group caused smaller involvement and area of DONFL in the temporal quadrant, which demonstrated that, by reducing the scope of ILMP in the temporal quadrant, the modified ILMP technology reduces the damage of the temporal retina.At the same time, this also verified that DONFL is an ILMP-related morphological change.
Previous studies have indicated that the occurrence of DONFL caused by ILM peeling is almost inevitable, 24,25 and can be detected about 1 month after ILMP. 26The incidence and severity of DONFL increased over time from 1 to 6 months after surgery, 26,27 and stabilized 6 months after the surgery. 16,28For a long time DONFL was thought to be an anatomical appearance that did not affect visual function and was associated primarily with dissociation rather than loss of nerve fibers. 29,30However, recent studies have suggested that DONFL may not be merely a modification of the nerve fiber layer, but may involve the GCL-IPL complex, 31 which led the researchers to wonder if DONFL appearance might cause some potential changes in visual function.Previous studies have shown that ILMP leads to the selective delay of recovery of the focal macular electroretinogram's b-wave. 32Some microperimetryrelated studies have shown that parafovea scotoma often appears after ILMP, and sometimes the distribution of these scotomas is similar to the distribution of DONFL. 33n fact, the present study does not prove that DONFL is merely a morphological change after surgery.More F I G U R E 5 Differences in the distribution patterns of CMDS in two ILM peeling groups were compared.(A) The mean radius of CMDS was 2.052 ± 0.458 mm in the traditional ILM peeling group, while the mean radius was 1.721 ± 0.235 mm in the modified ILM peeling group.There was a statistically significant difference between the two groups (p < .01).The radius of CMDS distribution of the temporal quadrant was significantly larger in the traditional group (p < .01).The pie stack diagram showed the distribution differences in the mean lesion radius within each quadrant of the two groups of patients.(B) The mean total area of CMDS was 0.385 ± 0.237 mm 2 in the traditional ILM peeling groupand 0.224 ± 0.110 mm 2 in the modified ILM peeling group.There is a significant difference between the two groups (p < .01).The area of CMDS distribution of the temporal quadrant was significantly larger in the traditional group (p < .01).The pie stack diagram showed the distribution differences of the mean lesion area within each quadrant of the two groups of patients.(C) The NFLDI of CMDS was 8.064 ± 3.207% in the traditional ILM peeling group, and 7.094 ± 3.062% in the modified ILM peeling group.There is no significant difference between the two groups (p > .05).The pie stack diagram showed the distribution differences of the mean DONFL index within each quadrant of the two groups of patients.
functional studies are needed to explore the mechanism of this appearance.
Nowadays, with a high anatomical closure rate of macular holes after surgery, reducing potential damage associated with surgery like "DONFL appearance" has become another important post-operational goal.Using the temporal inverted ILM flap technique, the incidence of DONFL was significantly reduced on B-scan OCT. 34However, our previous studies on en-face OCT using computer aids demonstrated that DONFL is mainly concentrated in the temporal quadrant. 16We speculate that this distribution pattern may be related to a thinner retina and The distributed heat map of the (A) radius, (B) area, and (C) NFLDI in each quadrant of postoperative CMDS for two groups.According to two different methods of traditional (T) and modified (M) ILM peeling groups, combined with the division of regions (N, S, T, I), all data was divided into eight categories.In the thermal diagram, the horizontal axis represented each type of information, the vertical axis represented the distribution of such parameters, and the numbers in the grid of the thermal diagram represented the frequency of (x, y) occurrence.
fewer large blood vessels at the temporal quadrant, resulting in less resistance to stretching and deformation of the retinal tissue.Therefore, in this study, we reduced the extent of temporal ILMP rather than nasal ILMP, which was followed by smaller involvement and a smaller area in the temporal quadrant DONFL.We speculate that modified ILMP may be a potential technique to protect visual function after surgery.
How to conduct a qualitative characterization of the DONFL an unresolved clinical problem.Previous studies on the development of DONFL have focused on changes in the number of inner retinal dimpling on B-scan OCT and the number of CMDS on en face OCT, [35][36][37][38] divided the changes in the inner retina after ILM peeling into CMDS with intact GC-IPL and CMDS with evident localized defects in the underlying GC-IPL based on en face OCT.A novel system using the LAT algorithm was developed for the detection and location of the DONFL on en-face OCT images.The system could automatically calculate the area of CMDS and calculate its percentage in each subregion after excluding the pixels on the macular hole, and an analysis report would be then exported.
This study exists some limitations: (1) We mainly studied the characters of CMDS on en-face OCT images, and the depth and volume of DONFL have not been computed.A three-dimensional analysis of DONFL should be performed.(2) This is a monocentric study, and the sample of eyes that underwent modified ILMP was relatively small.(3) This study compared the different appearances of DONFL resulting from the two surgical procedures but lacked corresponding functional analysis.Further studies with more functional indicators will be needed to better explain the mechanics of DONFL to guide d the improvement of surgery.This is the first study to accurately quantify and compare the DONFL appearance secondary to modified ILMP technology and traditional ILMP technology using an automated system.The results suggest that modified ILMP technology can reduce the extent of DONFL involvement in the whole macular area and reduce the involvement of the temporal retina, especially.Those results may provide a clinical research basis for the improvement of surgical methods.

CONCLUSIONS
The system proposed in this study can effectively and accurately quantify DONFL appearance in the macular area.Modified ILMP technology can reduce the extent of DONFL involvement in the macular area and reduce the involvement of the temporal retina.This study will provide

C O N F L I C T O F I N T E R E S T S TAT E M E N T
The author declares no conflict of interest.

F U N D I N G I N F O R M AT I O N
The sponsor or funding organization had no role in the design or conduct of this research.

D ATA AVA I L A B I L I T Y S TAT E M E N T
The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

F I G U R E 1
Representative images for the traditional ILM peeling surgery (A-F) and the modified ILM peeling surgery (D-L).(A-F) The images were from the right eye of a 62-year-old woman who was treated with vitrectomy and traditional ILM peeling for an idiopathic macular hole.(A-C) Preoperative fundus photography and B-scan OCT image showed a macular hole but not concentric macular dark spots (CMDS).(D) The ILM around the macular hole was removed after indocyanine green staining.(E) Arcuate dark striae distributed along the direction of nerve fibers were observed on en face at 6 months postoperatively.(F) Dimples of the inner retina were observed on B-scan OCT at 6 months postoperatively.(G-L) The images were from the left eye of a 56-year-old woman who was treated with vitrectomy and traditional ILM peeling for an idiopathic macular hole, and dimples of the inner retina were observed on B-scan OCT at 6 months postoperatively (L).

F I G U R E 2
Workflow of image processing and automatic measurement of CMDS.En-face OCT images of the superficial macular region of the retina were obtained with the RTVue XR OCT device self-contained 3D reconstruction software.Automatically measurement of CMDS in the temporal, nasal, superior, and inferior four quadrants, respectively.

F I G U R E 3
Scatter plot of patient age and the length of the ocular axis.According to the two different methods of offset stripping and traditional stripping, all data is divided into two categories.The corresponding (x, y) values for each point in the scatter plot are, where x represents the age of the object and y represents the eye distance information of the object.The labels in the figure indicate the maximum and minimum values, and provide an average line.

F I G U R E 4
The three-dimensional (3D) reconstruction rendering of the macular hole before and after surgery.(A, B) The 3D reconstruction rendering of macular hole.(C, D) The 3D reconstruction rendering of a patient that underwent the traditional ILM peeling surgery.(E, F) The 3D reconstruction rendering of a patient that underwent the modified ILM peeling surgery.TA B L E 2 The mean distribution parameter assessment of DONFL.

a
clinical research basis for the improvement of macular hole surgery.A U T H O R C O N T R I B U T I O NCYQ, LXX, and SLJ had full access to all the data in the study and will take responsibility for the integrity of the data and the accuracy of the data analysis.Study concept and design: CYQ, LXX, SLJ.Acquisition, analysis, or interpretation of data: YX, WSL, HSC, YJF, CH, ZSA, WCX, JSS, MJB, SLJ, LXX, CYQ.Drafting of the manuscript: YX, WSL, HSC, CYQ.Critical revision of the manuscript for important intellectual content: CYQ, LXX, SLJ.Study supervision: CYQ, LXX, SLJ.A C K N O W L E D G E M E N T SThis work was supported by the National Natural Science Foundation of China (81700884), Zhejiang Public Welfare Technology Application Project (LGF21H120005), and Wenzhou Scientific Research Project (Y20190627).

of the retina Characteristics Temporal Nasal Superior Inferior p Value Radius of involvement (mm)
Distribution parameter assessment of DONFL.
TA B L E 3 †One-way ANOVA; *Two independent-sample t-test.