Quantitative assessment of the corneal subbasal nerve plexus in children with neurofibromatosis type 1 and optic pathway glioma using in vivo confocal microscopy

Abstract

To quantitatively evaluate corneal subbasal nerve plexus (SNP) in children with neurofibromatosis type 1 (NF1) and optic pathway glioma using in vivo confocal microscopy (IVCM), comparing semi-automated and automated image analysis methods. In this prospective observational case series, 13 eyes of 7 children with NF1 and optic pathway glioma and 10 eyes of 6 age-matched healthy controls underwent IVCM imaging. Three high-quality central corneal images per eye were analyzed using NeuronJ (semi-automated) and ACCMetrics (automated) software. Corneal nerve fiber length (CNFL), nerve fiber density (CNFD), branch density, fiber area, fiber width, and fractal dimension were quantified. Group comparisons used t-tests and linear mixed modeling, accounting for inter-eye correlation. Agreement between methods was assessed with Bland–Altman analysis. NF1 eyes showed significantly greater CNFL on NeuronJ (27.13 ± 5.17 vs. 22.70 ± 4.98 mm/mm²; P = 0.042) and higher CNFD on ACCMetrics (27.32 ± 10.64 vs. 20.83 ± 7.98 fibers/mm²; P = 0.038) compared with controls. Other ACCMetrics parameters (CNFL, CNBD, CTBD, CNFA, CNFW, CNFrD) were not significantly different. Bland–Altman analysis demonstrated poor agreement between methods, with ACCMetrics systematically underestimating CNFL relative to NeuronJ (bias = 8.85 mm/mm²; limits of agreement 2.20 to 15.50 mm/mm²) and increasing underestimation at higher CNFL values (r = 0.43, P < 0.001). Children with NF1 and optic pathway glioma exhibit increased corneal nerve fiber length and density, suggesting enhanced nerve proliferation or branching rather than hypertrophy. However, automated ACCMetrics underestimated CNFL relative to semi-automated NeuronJ, emphasizing the need for analytic standardization. Larger, longitudinal studies are required to confirm these findings and explore their potential as noninvasive biomarkers of NF1-associated neuropathy.

Introduction

Neurofibromatosis type 1 (NF1) is an autosomal dominant neurocutaneous disorder caused by mutations in the NF1 gene on chromosome 17, with complete penetrance and variable expression¹. The prevalence is 1 in 3,000 births and occurs equally among both genders and races¹. Ophthalmic findings such as optic pathway glioma, plexiform neurofibroma, and Lisch nodules are widely known¹. In recent years, further ocular manifestations have been recognised namely, choroidal abnormalities, which have become part of the diagnostic criteria² and retinal vascular abnormalities, reported in large cohorts³. Prominent corneal nerves are another occasionally reported finding^4–6, although its structural basis and clinical implications remain poorly defined.

The corneal subbasal nerve plexus (SNP) is a dense network of thin sensory fibers located just under the basal epithelial cells and above Bowman layer⁴. These unmyelinated C-fibers and thinly myelinated Aδ-fibers are essential for corneal sensation, epithelial homeostasis, and ocular surface integrity^7,8. Quantitative alterations of the SNP have been demonstrated in systemic disease such as diabetes⁹, multiple sclerosis¹⁰, Parkinson disease¹¹, Alzheimer disease¹²,and post-viral condition such as COVID¹³, confirming the theory of the cornea as an noninvasive biomarker of systemic small fiber disease⁸. Additionally, chemotherapy-induced peripheral neuropathy has been associated with alterations of the corneal sub-basal nerve plexus detectable by in vivo confocal microscopy (IVCM)¹⁴.

IVCM is a high-resolution imaging technique that enables direct visualization of the SNP and has been validated as a marker for small fiber neuropathy in both adults and children^9,15. Quantitative analysis of IVCM images can be performed using semi-automated software such as NeuronJ or fully automated tools such as ACCMetrics, both of which have been validated^16–18.

Although enlarged corneal nerves in NF1 have been described clinically⁵ quantitative studies of SNP morphology in children with NF1 are scarce. In this study, we used both semi-automated and automated image analysis methods to quantify SNP architecture in children with NF1 and optic pathway glioma compared with age-matched healthy controls.

Materials and methods

Patients

This was a prospective observational case series conducted at Farabi Eye Hospital, Tehran, Iran. Thirteen eyes of seven NF1 patients with optic nerve glioma who were referred to neuro-ophthalmology clinic and 10 eyes of 6 healthy age-matched controls were included. The protocol of the study was in accordance with the declaration of Helsinki. Ethics Committee of Tehran University of Medical Science confirmed the study design.Written informed consent was obtained from the parents or legal guardians of all participants after a detailed explanation of the study procedures.

Inclusion criteria for NF1 patients were: (1) confirmed diagnosis of NF1 based on at least two NIH diagnostic criteria¹⁹, and (2) cooperative enough to undergo IVCM imaging. Exclusion criteria were history of corneal opacity, previous ocular trauma or surgery, contact lens wear, diabetes or other systemic diseases known to affect corneal nerves, and poor-quality IVCM images.

In vivo confocal microscopy

Heidelberg Retina Tomograph 3 with the Rostock Cornea Module (Heidelberg Engineering, Heidelberg, Germany) was performed for all patients. This device uses a 670-nm red wavelength diode laser source and has an immersion lens with a magnification of 63X. Each image represents an area of 400 × 400 μm equivalent to 384 × 384 pixels. Briefly, polymethylmethacrylate cap (Tomo-Cap; Heidelberg) was scaled foreside the cornea module, after covering the lens with 2.0 mg/g Carbomer, (Visicotears^®, Alcon, Fort Worth, Texas, United States). Tetracaine 0.5% ophthalmic drop (Anestocaine; Sinadarou, Tehran, Iran) was used for topical anesthesia. The cornea module was brought to touch the cornea. The section scan mode was used to capture image at the desired depth. Images were captured sequentially through the central cornea with special attention to the subbasal area. A total of 50–60 frames per cornea were captured. Three non-overlapping images with the highest quality, the least artifact with clear delineation of nerves from the background were selected for image analysis. Two independent observers analyzed all images. Identity of the participants was unknown for the observers. The average values of 3 images from the two observers were used for analysis.

Image processing

Semi-automated nerve tracing

Coded JPEG images were used for semi-automated nerve tracing using Neuron J plugin (https://imagescience.org/meijering/software/neuronj/) of Image J software (available in the public domain at http://imagej.nih.gov/ij/, 1997–2012; version 1.45s, Rasband, W.S., ImageJ; National Institutes of Health, Bethesda, MD) as described, previously^20,21. After tracing the nerves, the program displays the total length of all nerve fibers in pixels for each image. This value was then converted to micrometer with a coefficient of 1.0417 (400 μm of each image was equivalent to 384 pixels). Neuron J corneal nerve fiber length (njCNFL) represents mean value of corneal nerve fiber length measured by Neuron J for each eye.

Automated nerve tracing

ACCMetrics (MA Dabbah, Imaging Science and Biomedical Engineering, Manchester, UK) was used for fully automated nerve tracing and analysis of confocal images as previously described^18,21,22. ACCMetrics is an image analysis software that allows ​automatic quantification of nerve fiber metrics from single or multiple confocal images and is openly accessible through the University of Manchester portal.(The University of Manchester. Early Neuropathy Assessment (ENA) software. Available at: https://sites.manchester.ac.uk/ccm-image-analysis/). ACCmetric distinguishes nerve fiber from surrounding texture such as artifacts and connective tissue. A neural network classifierand a dual-model feature descriptor are employed for this purpose. The detected nerve fiber is used to specify the branching point and endpoints. All of these nervous components organize a connectivity map. This map includes main and branched fibers (Fig. 1) that are quantified as different variables which are summarized in Table 1.

Fig. 1.

In vivo confocal microscopy (IVCM) images of the corneal subbasal nerve plexus in a child with Neurofibromatosis type 1 (NF1) and in a healthy control. A and D show raw IVCM images from an NF1 patient and a control eye, respectively. B and E display the same images after semi-automated tracing with NeuronJ, while C and F show automated analysis with ACCMetrics. In the traced images, main nerves are highlighted in red, branches in blue, and branching points in green. Compared with the control, the NF1 eye demonstrates a greater number and length of visible nerve fibers, consistent with the quantitative findings.

Table 1.

Variables of ACCMetric.

Varibles	Abbreviation	Definition	Unit
Corneal nerve fiber density	CNFD	Total number of nerves per unit area	n/mm2
Corneal nerve branch density	CNBD	The total number of branches derived from the main nerves per unit area	n/mm2
Corneal nerve fiber length	CNFL	The length of all main and minor nerves per unit area	mm/mm2
Corneal nerve total branch density	CTBD	The total number of branches per unit area	n/mm2
Corneal nerve fiber area	CNFA	The total nerve fiber area per unit area	mm2/mm2
Corneal nerve fiber width	CNFW	The mean nerve fiber width per unit area	mm/mm2
Corneal nerve fractal dimension	CNFrD	a degree of the architectural complexity of corneal nerves16,17

Statistical analysis

SPSS statistical software, version 25 (SPSS Inc., Chicago, Illinois, USA), was used for statistical analysis. Continoues varibles were expressed as mean ± standard deviation (SD). Independednt sample t-test was used to compare continuous variables, and Fisher’s exact test was applied for categorical data. Linear mixed modeling was applied for the comparison between both groups. Determining a fixed effect and a random effect in this modeling nullify the confounding effect of inter eye correlation arising from incorporating both eyes of some cases in this study. A p-value < 0.05 was considered significant.

Results

Of 12 NF1 patients recruited, 5 (10 eyes) were excluded due to poor cooperation or image quality. Thus, 13 eyes of 7 NF1 patients and 10 eyes of 6 healthy controls were analyzed. The mean age of NF1 and control groups was comparable (7.0 vs. 6.8 years, P = 0.72). Males constituted 61.5% of the NF1 group and 70% of controls (P = 0.51).

Semi-automated analysis

Using the NeuronJ method, NF1 eyes demonstrated a significantly higher njCNFL compared with controls (27.13 ± 5.17 vs. 22.70 ± 4.98 mm/mm²; mean difference 4.43 mm/mm²; P = 0.042). In contrast, the mean segment length of individual fibers did not differ significantly between groups (172.24 ± 32.30 μm vs. 176.57 ± 47.43 μm; P = 0.781) (Table 2).

Table 2.

Quantitative parameters of the corneal Subbasal nerve plexus in NF1 patients and healthy controls, assessed by neuronj and ACCMetrics.

Varible	Method	NF1 eyes	Control eyes	P value
njCNFL (mm/mm2)	Neuron J	27.13 ± 5.17	22.70 ± 4.98	0.042
Mean lenghth (µm)	Neuron J	172.24 ± 32.30	176.57 ± 47.43	0.781
CNFD (n/mm2)	ACCmetrics	27.32 ± 10.64	20.83 ± 7.98	0.038
CNBD (n/mm2)	ACCmetrics	29.89 ± 17.23	30.31 ± 16.54	0.915
CNFL (mm/mm2)	ACCmetrics	17.19 ± 3.72	15.26 ± 3.50	0.154
CTBD (n/mm2)	ACCmetrics	53.20 ± 17.09	48.39 ± 26.00	0.601
CNFA (mm2/mm2)	ACCmetrics	0.0075 ± 0.0018	0.0074 ± 0.0021	0.869
CNFW (mm/mm2)	ACCmetrics	0.0206 ± 0.0012	0.0215 ± 0019	0.239
CNFrD	ACCmetrics	1.49 ± 0.03	1.48 ± 0.03	0.629

Automated analysis

Automated image analysis confirmed higher CNFD in NF1 eyes compared with controls (27.32 ± 10.64 vs. 20.83 ± 7.98 fibers/mm²; mean difference 6.49 fibers/mm²; P = 0.038). However, other key parameters derived from ACCMetrics including CNFL, branch density (CNBD and CTBD), fiber area (CNFA), fiber width (CNFW), and fractal dimension (CNFrD), were not significantly different between the two groups (Table 2).

Agreement between methods

Bland–Altman analysis was used to assess agreement between methods (difference defined as NeuronJ − ACCMetrics). The mean difference (bias) was 8.85 mm/mm² (one-sample t-test vs. 0: P < 0.001), indicating that ACCMetrics systematically underestimates CNFL relative to NeuronJ. The limits of agreement were wide, ranging from 2.20 mm/mm² (95% CI, − 0.34 to 4.74) to 15.50 mm/mm² (95% CI, 12.96 to 18.05). The difference increased with the magnitude of the measurement (proportional bias: r = 0.43, P < 0.001), consistent with larger underestimation at higher CNFL values (Fig. 2). We also showed an increase in the bias as the magnitude of the measurement increases. The mean differences were correlated to the magnitudes of the measurement (r = 0.43 and P < 0.001; Fig. 2).

Fig. 2.

Bland–Altman plot comparing semi-automated (NeuronJ) and automated (ACCMetrics) measurements of corneal nerve fiber length (CNFL). The solid horizontal line represents the mean difference (bias = 8.85 mm/mm²), while the dashed horizontal lines indicate the 95% limits of agreement (2.20 to 15.50 mm/mm²) with corresponding 95% confidence intervals. The purple dashed regression line illustrates a significant proportional bias (r = 0.43, P < 0.001), showing that ACCMetrics systematically underestimated CNFL relative to NeuronJ, with larger discrepancies at higher CNFL values. Table 2. Quantitative parameters of the corneal subbasal nerve plexus in NF1 patients and healthy controls, assessed by NeuronJ and ACCMetrics. Values are presented as mean ± standard deviation (SD). Independent sample t-test was used for group comparisons. Linear mixed modeling was applied to adjust for inter-eye correlation. P-values < 0.05 were considered statistically significant. CI = confidence interval; SD = standard deviation; NF1 = Neurofibromatosis type 1; CNFL = corneal nerve fiber length; CNFD = corneal nerve fiber density; CNBD = corneal nerve branch density; CTBD = corneal total branch density; CNFA = corneal nerve fiber area; CNFW = corneal nerve fiber width; CNFrD = corneal nerve fractal dimension.

Discussion

In this study, we quantitatively assessed corneal subbasal nerves in children with NF1 and optic pathway glioma using IVCM. Both semi-automated (NeuronJ) and automated (ACCMetrics) methods were applied. NF1 eyes demonstrated significantly greater corneal nerve fiber length and density compared with age-matched healthy controls. Other morphometric parameters, including branch density, fiber width, and fractal dimension, did not differ significantly between groups.

The cornea is richly innervated by both unmyelinated C fibers and thinly myelinated Aδ fibers (0.2–5 μm in diameter), which originate from the trigeminal ganglion²³. Nerve trunks enter the cornea radially through the mid-stroma, lose their myelin sheaths within 1 mm, and course parallel to stromal lamellae, wrapped by Schwann cells^24–26. Stromal nerves penetrate Bowman layer and branch extensively, giving rise to the SNP, which lies just beneath basal epithelial cells²⁷. The SNP is composed predominantly of C fibers, which terminate as free nerve endings at the epithelial surface²⁸. Because of its accessibility and dense architecture, the SNP has been widely studied by IVCM in ocular and systemic diseases including keratoconus²⁹, dry eye³⁰, herpes simplex keratitis³¹, Acanthamoeba and fungal keratitis³², corneal dystrophies^33,34, diabetes, and multiple neuropathies³⁵.

Clinically visible corneal nerves can result from either true nerve enlargement or increased nerve reflectivity^5,28,36. They have been reported in keratoconus, ichthyosis, Fuchs’ dystrophy, and congenital glaucoma³⁷. By contrast, genuine nerve enlargement has been described in Multiple endocrine neoplasia 2 A/2B, leprosy, Refsum disease, Acanthamoeba perineuritis, and multiple myeloma⁵. In MEN2B, for example, mid-stromal nerves may become five to six times thicker, with subbasal nerves showing increased density, looping, and nodularity³⁶. Histology has attributed these changes to Schwann cell proliferation, axonal hyperplasia, and increased myelination³⁸.

NF1 is a neurocutaneous disorder affecting neural crest-derived tissues and Schwann cells, with hallmark features such as café-au-lait macules, Lisch nodules, neurofibromas, and optic pathway gliomas.1 Approximately one-quarter of NF1 patients exhibit prominent corneal nerves clinically, attributed mainly to enlarged stromal nerves⁵. Myelination of corneal nerves has also been reported in NF1³⁹. However, the structural basis of these findings has remained uncertain.

Our study is the first, to our knowledge, to quantify SNP morphology in children with NF1. We found increased nerve fiber length and density but no increase in fiber thickness or complexity. This suggests greater nerve proliferation or branching rather than hypertrophy. The absence of increased CNFA, CNFW, or CNFrD indicates that subbasal nerves in NF1 are not thicker or more complex, consistent with the hypothesis that prominent nerves observed at the slit lamp primarily reflect stromal rather than subbasal alterations. Javadi Schwann cell proliferation and aberrant axonal support are known in NF1, but our findings argue against substantial replacement of small unmyelinated C fibers with larger Aδ fibers within the SNP.

The discrepancy between NeuronJ and ACCMetrics measurements deserves emphasis. ACCMetrics employs a higher contrast threshold to minimize recognition of artifacts, which may lead to underdetection of thin, low-contrast fibers particularly relevant in pediatric cohorts with subtle nerve changes. This explains why NeuronJ detected longer total nerve length while ACCMetrics did not, although both confirmed higher CNFD in NF1. Similar underestimation of CNFL by ACCMetrics compared with semi-automated or manual methods has been reported previously^16,21,40. Raasing et al.¹⁸ (2023) extended these observations in a sarcoidosis cohort, comparing manual (CCMetrics), semi-automated (NeuronJ), and automated (ACCMetrics) analyses. They reported strong correlations among methods but consistent underestimation of CNFL by ACCMetrics, particularly when nerve alterations were subtle. Their results mirror our findings, underscoring that automated analysis, while efficient, may miss fine low-contrast fibers. Bland–Altman analysis in our study further confirmed poor agreement between methods, with ACCMetrics systematically underestimating CNFL and showing increasing bias at higher values.

Although limited by small sample size, our findings suggest possible clinical relevance. They support the idea that the cornea may serve as a noninvasive biomarker of small fiber pathology in NF1, with IVCM offering complementary information to neuro-ophthalmic evaluation. Increased subbasal nerve parameters may indicate that clinically visible corneal nerves in NF1 reflect true structural changes, though their functional significance remains unclear. Larger studies are needed to confirm these observations and explore potential correlations with ocular surface health and longitudinal monitoring in children with NF1.

This study has several limitations. The sample size was small, reflecting the rarity of pediatric NF1 with optic pathway glioma. Only NF1 patients with glioma were included, limiting generalizability to NF1 without glioma. We evaluated only the SNP, not stromal nerves, which are often clinically prominent in NF1. Furthermore, automated analysis underestimated nerve length compared with NeuronJ, emphasizing the need for standardization of analytic methods.

In conclusion, children with NF1 and optic pathway glioma exhibited increased subbasal nerve fiber length and density without corresponding increases in fiber thickness or architectural complexity. These findings suggest enhanced nerve proliferation or branching rather than hypertrophy and highlight the potential of IVCM as a noninvasive biomarker of NF1-associated neuropathy. Larger, longitudinal studies are needed to validate these results and explore correlations with clinical outcomes.

Author contributions

H.G. conceptualized and designed the study, acquired and analyzed the core dataset, and drafted the initial manuscript.G.L. contributed to data collection and analysis, created the figures and tables, and provided critical revisions to the manuscript.M.F. supervised the project, validated the data, provided methodological oversight, and finalized the manuscript with substantive edits.

Data availability

The datasets are available from the corresponding author upon request.

Declarations

Competing interests

The authors declare no competing interests.