Correlation between AI-measured lacquer cracks extension and development of myopic choroidal neovascularization

To investigate the correlation between the AI-measured area of the lacquer cracks (LC) at their first detection and the occurrence of a choroidal neovascularization (CNV) during the follow-up in patients affected by pathologic myopia. Secondary outcome was the detection of a correlation between the time to onset of CNV with both baseline LC area and LC area increase during follow-up. Optical coherence tomography (OCT) acquisitions of patients diagnosed with LC were retrospectively analysed. The study population was divided in a CNV group (showing the documented onset of a CNV) and a n-CNV group (no CNV development during follow-up). LC area was measured using MatLab software after the application of a customized method for LC segmentation on infrared (IR) enface images. Forty-five (45) patients with a mean follow-up of 4.9 ± 1.5 years were included. LC area at baseline was 2.82 ± 0.54 mm2 and 1.70 ± 0.49 mm2 in CNV (20 patients) and n-CNV group (25 patients) group respectively (p < 0.001). LC area increase was significantly higher in CNV group (p < 0.001). Time to onset of CNV was linearly correlated with both LC area at baseline (p = 0.006) and LC area increase (p < 0.001). Myopic CNV development is associated with lager LC areas and higher LC area increase during time. Earlier CNV onset is inversely correlated with LC area and LC area increase.


INTRODUCTION
Pathological myopia is defined as the coexistence of high myopia and anatomical characteristics predisposing to atrophic, traction, or neovascular myopic maculopathy, such as posterior staphyloma and elongation of the axial length [1].Bulbar elongation is responsible not only for anomalies in posterior pole retinal adherence (determining complications such as dome shape macula, full thickness macular hole, myopic foveoschisis and posterior pole detachment) but also for choroidal and retinal thinning, that are associated to atrophic degeneration and neovascular membranes formation in pathologic myopia.Myopic choroidal neovascularization (CNV) is the second most common cause of CNV after agerelated macular degeneration (AMD).It occurs in 5 to 11% of patients with pathological myopia, being responsible for a high socioeconomic impact and determining a high degree of disability in everyday life activities in these patients.Its development has been directly correlated to the relative height of posterior staphyloma and inversely correlated to choroidal thickness (CT) [2].In fact, according to the mechanical theory of myopic CNV formation, mechanical choroidal and scleral strain induced by elongation of the eyeball produces zones of fragility of the RPE-Bruch membrane complex facilitating retinal ingrowth of neovessels from the underlying choriocapillaris [3].Lacquer cracks (LC) are linear atrophic lesions believed to be the manifestation of acute ruptures in the RPE-Bruch's membrane-choriocapillaris complex caused by the stretching effect of bulbar elongation [4].They occur at the posterior pole in 4.2% to 15.7% of highly myopic eyes, especially in patients with posterior staphyloma, male sex and younger age [5].The higher prevalence of LC both in the affected and fellow eye of patients developing a myopic CNV is among the elements supporting the mechanical theory of CNV formation [6].In a large series from Ohno-Matsui et al. [7], myopic CNV developed in around 10% of highly myopic eyes and specifically in 3.7% of eyes with diffuse chorioretinal atrophy, in 20.0% of eyes with patchy atrophy, and in 29.4% of eyes with LC.Nevertheless, the poor specificity of the association detected to the present date impairs the use of LCs as predictors of CNV development in pathologic myopia.Recent development of image processing and artificial intelligence (AI) techniques made it possible to extract unprecedented qualitative and quantitative information from imaging acquisitions, that have already shown surprising efficacy in unrevealing hidden characteristics of other retinal diseases [8,9].The purpose of this study was to investigate the presence and the relevance of a correlation between the AI-measured area of the LC at the moment of their first detection and the occurrence of a CNV during the follow-up in patients affected by pathologic myopia.Moreover, we aimed to find a correlation between the baseline LC area and the time to onset of CNV after LC first detection.

MATERIALS AND METHODS
Patients affected by pathologic myopia and showing LCs in at least one eye were retrospectively selected from databases of two specialized retina centres (Department of Ophthalmology of the University Paris Est Créteil and Department of Ophthalmology of IRCSS San Raffaele Milano) among records from March 2014 to August 2022.Pathologic myopia was defined as the presence of choroidal and retinal myopic degeneration in patients with an axial length>26 mm and a spherical equivalent (SE) < −6D [10][11][12].Additional inclusion criteria were availability of high quality (Q > 15) enface infrared (IR) acquisitions documenting the first appearance of LC (acquisitions before and after their appearance) and age older than 18 years.A minimum follow-up of 2 years was required for inclusion in n-CNV group.LC presence was assessed by three experienced graders (E.C., M.B. and S.F) based on the examination of high resolution (1536 × 1536 pixels) 30°x 30°f ovea-centred enface IR and late phase ICGA images [13][14][15].All images from both centres were acquired using Spectralis spectral domain (SD) OCT (Heidelberg Engineering, Heidelberg, Germany).LC were defined as hyperreflective linear lesions, in and around the macula, on a 30°x 30°IR acquisition.LC presence was confirmed with late-phase ICGA imaging when available, and they were identified as hypocianescent linear lesions due to choroidal filling defects [16].Exclusion criteria were diffuse choroidal atrophy at the moment of appearance of LC preventing accurate full evaluation of the LC area, concomitance of other retinal diseases and interval between follow-up visits >6 months.The development of a myopic CNV during the follow-up was registered and the study population was divided in two groups: a CNV group, showing OCT documented onset of a CNV during the follow-up time, and a n-CNV group, in which no CNV developed according to available records.Relevant risk factors for the development of CNV in myopic patients were registered so to consider potential confounding factors.Considered factors included best corrected visual acuity (BCVA), spherical equivalent, axial length, absolute height of posterior staphyloma, CT, central macular thickness, the presence of patchy choroidal atrophy, familiarity for myopic CNV, previous cataract surgery [2,17], Axial length was derived from optical biometry calculations performed with IOL Master 700 or IOL Master 500 (Carl Zeiss Meditec AG, Jena, Germany) when available.The height of posterior staphyloma was calculated on 55°x 55°B-scan OCT images as the perpendicular distance between RPE-Bruch junction at 3000 μm superior, nasal, temporal and inferior to the fovea and a line drawn tangential to the foveal RPE-Bruch junction.The mean of the 4 distances expressed in absolute positive value was defined as absolute posterior staphyloma height, as previously described in the literature [2].CT was defined as the distance from the RPE line to the hyper refractive line posterior to the large vessel layers of the choroid.Subfoveal CT was measured on OCT B-scan images as the mean of manual measurements performed by 3 expert retina specialists.Mean of CT at 1500 μm from the fovea in the superior, nasal, inferior, and temporal sector was also calculated as parafoveal CT.The aim of the study was to investigate the presence and the relevance of a correlation between AImeasured baseline LC area and CNV formation in myopic maculopathy patients with LC.Secondary outcomes were the assessment of a correlation between LC area at baseline and time to CNV onset and the assessment of an association between LC area increase during follow up and time to CNV onset.For LC area calculation, IR images were first cropped using an opensource image processing software (ImageJ, NIH, Bethesda, MD) (Fig. 1A).LC segmentation was performed with MatLab software (Mathworks, Natick, MA) using a manual threshold of 185 and an adaptive threshold of 35% sensitivity (Fig. 1B).The adaptive threshold used was chosen as the one resulting in the best agreement with manual contouring performed by expert graders on ROC curve analysis.Erosion for disc shapes with radius = 2 was then performed to refine the segmentation.Validation of the segmentation from a human grader was necessary to ensure that segmented images matched the real shape and dimensions of the lacquer cracks as seen by human eye.Spurious and patch atrophy areas were excluded (Fig. 1C).Areas of the segmented regions was then calculated and converted to mm 2 (Fig. 1D).The difference between LC area at baseline and LC area at final follow-up acquisition was calculated and defined as Δ LC area (mm 2 ).The study was conducted in accordance with the tenets of the Declaration of Helsinki and with French and Italian legislation and was approved from the Ethics Committee of the Federation France Macula 2018-27.

Statistical analysis
Statistical analysis was conducted using SPSS v.26 (IBM SPSS statistics).Normal distribution of the quantitative variables was assessed with Shapiro-Wilk test and variables were described using mean and standard deviation.Qualitative variables were described with a number of cases over total and percentage.Comparison of quantitative variables was performed using Χ 2 test while qualitative variables were compared using two-tail T-test for independent samples.Linear regression was performed using CNV development during follow-up as a dependent variable and all quantitative variables resulting as statistically significant at univariate analysis as independent variables.Linear correlation between LC area and time to onset of CNV and between Δ LC area and time to onset of CNV was assessed using Spearman correlation.A p-value <0.05 was considered as statistically significant.

RESULTS
total of 45 patients was included with a mean follow up of 4.9 ± 1.5 years.Mean spherical equivalent was −9.9 ± 2.4 D, ranging from −6.25 to −12.75 D. Twenty (20) patients with a mean age of 66 ± 8.7 years developed a CNV during the follow-up (CNV group), with a mean time to onset of CNV of 10.4 ± 8.1 months and a mean follow up of 4.9 ± 1.7 years.Mean number of intravitreal injections (IVI) per person was 4.2 ± 2.6 and all 20 patients were treated with ranibizumab injections.Patients included in n-CNV group were 25 and were followed for 4.8 ± 1.3 years, which was not statistically different from follow up of CNV group (p = 0.751).No significant differences in terms of age and sex compositions were detected between the two groups (see Table 1).LC area at baseline was 2.82 ± 0.54 mm 2 and 1.70 ± 0.49 mm 2 in CNV and n-CNV group respectively (p < 0.001, B = 0.153) (see Figs. 2 and 3).LC area at baseline showed a significant linear correlation with time to onset of CNV (R = −0.87(CI from −0.94 to −0.69), p = 0.006, see Fig. 4).Similarly, the LC area increase during follow-up (Δ LC area) was    degree of correlation (R = 0.959, R = 0.918, F = 107.127).Moreover, LC area showed a positive correlation with Δ LC area (LC area increase during follow-up) (R = 0.62) and absolute staphyloma height (R = 0.67) (Fig. 4).Similarly, Δ LC area showed a significant positive correlation with LC area (R = 0.62) and absolute staphyloma height (R = 0.73).Both LC area and Δ LC area were negatively correlated with subfoveal CT and parafoveal CT (see Fig. 4).

DISCUSSION
Myopic macular degeneration accounts for up to 7.8% of causes of severe vision loss in European countries [18] and significantly impacts public health's costs and patients' quality of life, particularly as concerns vision-related activity limitation [19].
Myopic CNV accounts for a major part of critical VA loss cases in pathologic myopia and refinement of predictors of its onset may prompt indications for a closer follow-up in selected patients and help earlier identification of CNV.LC have been associated to higher risk of CNV in past literature, with up to 30% of LC patients developing a CNV within 3 years from identification [7].Ikuno et al. [20] found that 94% of CNV in myopic patients developed from LC regions with all CNV being of classic type.This supports the theory of the pathogenesis of myopic CNV being linked to areas of RPE fragility predisposing intraretinal migration and ingrowth of choroidal neovassels.Nevertheless, lack of characterization of the type of LC lesions conferring the highest risk for CNV onset impaired their specificity as a prognostic factor.Some attempts have been made to correlate morphology of the LCs to the risk of CNV, with conflicting results: while in some series no differences were found between linear and stellate LCs [20], others identified a higher risk associated to crossing LCs compared to linear ones [2].The advancements of technology made possible to overcome limitations given by shape-based classification allowing precise and automated measurements of LC area.Using this method, we identified LC area at baseline examination as being significantly associated to CNV onset on a 5-year follow-up retrospective investigation.Not only patients that developed a CNV during follow-up (CNV group) showed a higher LC area at baseline, but they were also characterized by a higher increase in LC area between first and last examination (Δ LC area), thus showing a progression of the pathological process over time.Moreover, both LC area and Δ LC area displayed an inverse linear correlation with time to CNV onset from baseline examination, meaning that larger and more progressing lesions are correlated to a faster development of CNV.Consistently with previous findings [2] and with the theory of LC being induced by bulbar elongation, LC area and LC progression were positively correlated with absolute staphyloma height and negatively correlated with subfoveal and parafoveal choroidal thickness.Interestingly, according to our findings, higher LC area at baseline was correlated to higher increase in LC area during the follow-up.Both features might be manifestations of a more active bulbar strain process, leading to choroidal thinning, higher staphyloma height and faster CNV development.Lastly, our findings confirmed lower subfoveal and parafoveal CT, higher absolute staphyloma height and patchy choroidal atrophy to be associated to CNV development.By contrast, patients that developed CNV did not show differences in terms of axial length, SE and BCVA at baseline compared to patients who did not develop CNV within 5 years from examination.The reported findings are limited by the retrospective nature of the study, in particular as concerns assumptions on time of onset of CNV.Nevertheless, the high performances of the linear regression model based on LC area, Δ LC area, subfoveal CT, parafoveal CT and absolute staphyloma height revealed that these parameters are likely to provide a good prediction of myopic eyes at higher risk for CNV development.Past literature [14,21] demonstrated high performances of N-IR imaging in LC detection (92.9% concordance with ICGA) and showed how hyperreflective lines on N-IR corresponded to hypocianescent lines on late-phase ICGA.
Considering the high reliability in LC measurement and detection showed in background literature [5,14,16,22] and the scope of our study, the higher generalizability of results obtained using IR images rather than ICGA made us choose IR as a method for LC area measurement.In fact, ICGA is not routinely performed in myopic patients due to invasiveness of the exam while IR enface imaging represents a practical non-invasive tool that could allow easy measurement in everyday clinical practice.In agreement with current literature and as highlighted in Fig. 2, LC profile shown by ICGA images in closely similar to the one shown by IR images.Consequently, even though ICGA remains the gold standard for LC detectability, IR can provide good performance il LC area progression calculation.To the best of our knowledge, it is the first report in literature to provide measurement of LC area and LC area progression and to highlight its importance as a prognostic factor in pathologic myopia.We found that larger LC areas and higher LC area increase during time are associated to CNV onset in pathologic myopia population.Moreover, the onset is faster in patients showing large LC areas at baseline and a large LC area progression at follow-up examinations.Lastly, we included both LC area and LC area increase in a high performing for CNV risk assessment.We hope that our findings might stimulate further prospective research to confirm our hypothesis and to provide personalized screening of CNV development in myopic population.

SUMMARY
What was known before • Myopic CNV is believed to be elicited from the formation of alterations in RPE-Bruch membrane complex.
• Lacquer cracks are associated to an increased risk of myopic CNV.

What this study adds
• Lacquer cracks area and its increase over time measured on infrared enface images are larger in patients developing a CNV.
• Lacquer crack area and its increase over time are inversely correlated to time to onset of CNV.
• Lacquer crack area is directly correlated to absolute staphyloma height and inversely correlated to choroidal thickness.

Fig. 1
Fig. 1 LC segmentation process.A Enface IR image.Details of pixels' luminance of a small area (blue rectangle) at the border of a LC are highlighted; B Segmentation based on manual and adaptive thresholding followed by erosion (as described in the Materials and methods section); C Exclusion of spurious areas; D Binarization of the segmented region for area calculation.IR infrared, LC lacquer cracks.

Fig. 2
Fig. 2 Evolution of LC in a patient from CNV group.A Baseline late-phase ICGA; B Baseline enface IR.LC profile is matching the one in A; C Fovea-crossing OCT B-scan at baseline; D Follow-up late-phase ICGA; E Follow-up enface IR; F Fovea-crossing OCT B-scan with CNV visible (red arrowhead); G Fluoresceine angiography showing CNV (red arrowhead).CNV onset occurred 4 months after the acquisition of image A and was treated with anti-VEGF therapy.Follow-up images shown (D-G) were acquired 2.2 years after baseline, and thus represent endfollow-up acquisitions.The enlargement of the LC area is visible between baseline and follow-up acquisitions, even in areas not involved by the CNV.CNV choroidal neovascularization, IR infrared, LC lacquer cracks, OCT optical coherence tomography.

Fig. 3
Fig.3Enface IR and OCT B scan of a patient who did not develop CNV within follow up period.Slight enlargement of the LC area is visible between baseline (A) and follow up (B) acquisitions.LC area is visibly lower than the one of patient in Fig.2, both in baseline and follow-up acquisitions.CNV choroidal neovascularization, IR infrared, LC lacquer cracks, OCT optical coherence tomography.

Fig. 4
Fig. 4 Upper graphs: Linear correlation between time to onset of CNV and LC area (left image) and Δ LC area (right image).Lower graph: Correlation matrix showing correlations between quantitative variables included in linear regression analysis.Positive correlations are coloured in blue while negative correlations are coloured in red.Corresponding R values are marked into each box.CNV choroidal neovascularization, LC lacquer cracks.

Table 1 .
Results of inferential statistic comparing the two study groups.
Results of quantitative variables are reported as mean and standard deviation (SD).BCVA best corrected visual acuity, CNV choroidal neovascularization, LC lacquer cracks, OR odds ratio, SE spherical equivalent.Statistically significant values are highlighted in bold.