The Utility of Corneal Nerve Fractal Dimension Analysis in Peripheral Neuropathies of Different Etiology

Abstract

Purpose

Quantification of corneal confocal microscopy (CCM) images has shown a significant reduction in corneal nerve fiber length (CNFL) in a range of peripheral neuropathies. We assessed whether corneal nerve fractal dimension (CNFrD) analysis, a novel metric to quantify the topological complexity of corneal subbasal nerves, can differentiate peripheral neuropathies of different etiology.

Methods

Ninety patients with peripheral neuropathy, including 29 with diabetic peripheral neuropathy (DPN), 34 with chronic inflammatory demyelinating polyneuropathy (CIDP), 13 with chemotherapy-induced peripheral neuropathy (CIPN), 14 with human immunodeficiency virus-associated sensory neuropathy (HIV-SN), and 20 healthy controls (HCs), underwent CCM for estimation of corneal nerve fiber density (CNFD), CNFL, corneal nerve branch density (CNBD), CNFrD, and CNFrD adjusted for CNFL (ACNFrD).

Results

In patients with DPN, CIDP, CIPN, or HIV-SN compared to HCs, CNFD (P = 0.004–0.0001) and CNFL (P = 0.05–0.0001) were significantly lower, with a further significant reduction among subgroups. CNFrD was significantly lower in patients with CIDP compared to HCs and patients with HIV-SN (P = 0.02–0.0009) and in patients with DPN compared to HCs and patients with HIV-SN, CIPN, or CIDP (P = 0.001–0.0001). ACNFrD was lower in patients with CIPN, CIDP, or DPN compared to HCs (P = 0.03–0.0001) and in patients with DPN compared to those with HIV-SN, CIPN, or CIDP (P = 0.01–0.005).

Conclusions

CNFrD can detect a distinct pattern of corneal nerve loss in patients with DPN or CIDP compared to those with CIPN or HIV-SN and controls.

Translational Relevance

Various peripheral neuropathies are characterized by a comparable degree of corneal nerve loss. Assessment of corneal nerve topology by CNFrD could be useful in differentiating neuropathies based on the pattern of loss.

Introduction

The prevalence, severity, and disability associated with different peripheral neuropathies vary by etiology and can pose a significant diagnostic and therapeutic challenge. The diagnosis of peripheral neuropathies has traditionally relied on identifying typical symptoms, neurological deficits with more detailed assessment using neurophysiology, quantitative sensory testing, autonomic function testing, and, if necessary, a nerve or skin biopsy.¹ CCM is a non-invasive ophthalmic imaging device which has been used to show reduced corneal subbasal nerves in DPN,^2,3 to evaluate the severity of diabetic neuropathy,^4–7 and to predict incident neuropathy.^8,9 Furthermore, CCM has been used to show significant improvement in corneal nerve morphology in several intervention studies.^10–15

More recently, CCM has been used to investigate other peripheral neuropathies.^16–22 Stettner et al.¹⁶ reported corneal nerve loss and increased immune cell infiltrates in patients with chronic inflammatory demyelinating polyneuropathy (CIDP). Kemp et al.¹⁷ reported a reduction in corneal nerve density and increased tortuosity in patients with human immunodeficiency virus-associated sensory neuropathy (HIV-SN). Ferdousi et al.¹⁸ found a decrease in the corneal nerve fiber length (CNFL) of patients with gastroesophageal cancer and an increase in patients with chemotherapy-induced peripheral neuropathy (CIPN). Corneal nerve loss has been demonstrated in a cohort of patients with transthyretin familial amyloid neuropathy²³ that was related to the severity of somatic and autonomic neuropathy, but most importantly the nerve loss could be quantified in all of the patients, avoiding the floor effect seen with sural nerve action potential and intra-epidermal nerve fiber density. More recently, we have identified corneal nerve loss in patients with Fabry disease,²⁴ Friedreich's ataxia,¹⁹ and sarcoidosis²⁵ and have found significant correlations with neurological disability.

Given that corneal nerve loss occurs in such a wide range of peripheral neuropathies, this raises a question regarding the clinical utility of CCM for differentiating neuropathies of differing etiology. In a study of patients with type 1 diabetes, in addition to the established measures of corneal nerve morphology we have shown that corneal nerve fractal dimension (CNFrD) may allow the identification of those with DPN.²⁶ Fractal dimension²⁷ is a quantitative estimate of the topological complexity of an image feature and has been employed to describe structural alterations in the brain of patients with neurodegenerative,^28,29 neurodevelopmental,³⁰ and psychiatric disorders.³¹ Fractal analysis allows an assessment of the topology of the main nerve fiber and branches, which, we propose, may differ according to the etiology of peripheral neuropathy. In this study, we applied fractal dimension analysis of corneal subbasal nerves in patients with metabolic (DPN), inflammatory (CIDP), toxic (CIPN), and infectious (HIV-SN) peripheral neuropathies.

Methods

Study Subjects

CCM images from 29 subjects with type 1 diabetes mellitus and DPN (58.5 ± 12 years old), 34 patients with CIDP (59.2 ± 14.8 years old), 13 patients with CIPN (63.8 ± 10 years old), 14 patients with HIV-SN (57.7 ± 7.8 years old), and 20 age-matched healthy controls (HCs) (58.5 ± 12 years old) were used in this study. The data presented here constitute a secondary analysis of already published data,^16–18,26 but this study utilized only the data for patients with clinically diagnosed peripheral neuropathy. This study adhered to the Tenets of the declaration of Helsinki, and ethical approval was obtained by all participating institutions. Informed written consent was obtained from all subjects prior to participation. Exclusion criteria were any cause of neuropathy other than the primary cause of neuropathy (deficiency of vitamin B₁₂ or folate; autoimmune infectious; rheumatological disease) or having corneal dystrophies, active anterior eye infections, using contact lens on a regular basis, or having had recent (<1 year) refractive or cataract surgery.

Corneal Confocal Microscopy

All study participants underwent CCM with a Heidelberg Retinal Tomograph III Rostock Cornea Module (Heidelberg Engineering, Heidelberg, Germany). This device uses a 670-nm wavelength helium neon diode laser, which is a class I laser and therefore does not pose any ocular safety hazard. A 63× objective lens with a numerical aperture of 0.9 and a working distance, relative to the applanating cap (TomoCap, Heidelberg Engineering), of 0.0 to 3.0 mm was used. The size of each two-dimensional image produced was 384 × 384 pixels, corresponding to 400 µm × 400 µm and equivalent to 1.04 µm/pixel. To perform the CCM examination, local anesthetic (0.4% benoxinate hydrochloride; Chauvin Pharmaceuticals, Chefaro, UK) was used to anesthetize each eye, and Viscotears (carbomer 980, 0.2%; Novartis UK, London, UK) was used as the coupling agent between the cornea and the applanating cap. All subjects were asked to fixate on an outer fixation light throughout the CCM scan, and an external CCD camera was used to position the applanating cap correctly onto the central cornea. Images were acquired using the “section” mode in the Heidelberg Eye Explorer software. Based on depth, contrast, and focus position, six images per subject (three per eye) from the central subbasal nerve plexus were selected for analysis. CCM image examples from a HC and from participants with HIV, CIPN, CIDP, and T1DM are presented in Figures 1A to 1E.

Figure 1.

(Top row) Original CCM images from a HC (A) and from patients with HIV-SN (B), CIPN (C), CIDP (D), and DPN (E). (Middle row) Corresponding images analyzed with CCMetrics for estimation of CNFD (red), CNBD (green dots), and CNFL (red and blue) in HCs (F) and in patients with HIV-SN (G), CIPN (H), CIDP (I), or DPN (J). (Bottom row) Images analyzed for estimation of the novel metric CNFrD in HCs (CNFrD = 1.55) (K), HIV-SN (CNFrD = 1.5) (L), CIPN (CNFrD = 1.51) (M), CIDP (CNFrD = 1.45) (N), and DPN (CNFrD = 1.41) (O). Although CNFD and CNFL were significantly reduced in groups of patients compared to HCs, the underlying morphology assessed by CNFrD also appears significantly altered compared to HCs and among groups of patients.

Image Analysis

In total, 660 CCM images from both eyes were analyzed independently using manual semiautomated, purpose-designed CCM image segmentation software (CCMetrics; M.A. Dabbah, X. Chen, J. Graham, and R.A. Malik, University of Manchester, Manchester, UK) (Figs. 1F–1J) and the automated version (ACCMetrics).³² CCMetrics is a validated CCM image segmentation algorithm that allows measurement of corneal nerve fiber density (CNFD) (fibers/mm²), the number of main nerve fibers per image divided by the area of the image; corneal nerve branch density (CNBD) (branches/mm²), the number of main nerve branches; CNFL (mm/mm²), the sum of the length of all nerves per image (main nerves and branches). ACCMetrics was used for the estimation of CNFrD as previously described.²⁶ For CNFrD estimation, all original CCM images (Fig. 2A) were resized to 512 × 512 pixels, as this was the image resolution used to train our machine learning-based nerve fiber detection mode.³² CNFrD was calculated based on the binary image representing the detected nerve fibers using our previously developed machine learning approach,³² as shown in Figure 2B. CNFrD measures nerve complexity as the ratio of the change in detail to the change in scale in a CCM image. Box counting is a widely used and acknowledged method for fractal analysis calculation,³³ and it was applied in our study. This technique uses various sizes of n × n boxes (in our study, n = 1, 2, 4, …, 512, as 512 was the image width) to split the image into small patches. For each of the box sizes, the total number of boxes, Y_i (i is the index of the different box sizes), that contain nerve fibers are recorded. As shown in Figure 2C, the logarithm of Y against the logarithm of n is then plotted. A first-order polynomial (i.e., red line in Fig. 2C) is fitted to these points using the method of least squares, where the coefficient of the line fitting (i.e., slope of the line, denoted as C) is calculated. Subsequently, –C is used as the fractal dimension value of a given CCM image (i.e., CNFrD). The CNFrD value is unitless, as it is a ratio, and increases when the number of counted small boxes increases, indicating a more complicated structure (e.g., higher CNFrD in a healthy control subject). In contrast, fewer, shorter, or disrupted nerve fibers result in a lower CNFrD value that reflects altered morphology (Figs. 1K–1O). To adjust for the severity of corneal nerve fiber loss, a ratio of CNFL/CNFrD was calculated (CNFrD adjusted for CNFL, or ACNFrD), which is also unitless. The averaged result of six images per patient was used for CNFD, CNBD, CNFL, CNFrD, and ACNFrD.

Figure 2.

(A) Original CMM image, (B) detected nerve fibers using a machine learning method,³² and (C) CNFrD calculation using the box counting method.

Statistical Analysis

GraphPad Prism 8 (GraphPad Software, San Diego, CA) was used for the statistical analysis and graphic illustrations. A Shapiro–Wilk test was used to assess data for normality (P < 0.05). Data were confirmed to follow a normal distribution, and one-way analysis of variance (post hoc Bonferroni–Sidak test) was used to compare the results among participant groups and controls, respectively. All data are expressed as mean ± standard deviation, and a P < 0.05 was considered statistically significant.

Results

Demographics

There was no significant difference in age among the different groups of patients except for patients with CIPN, who were slightly older (P = 0.06) compared to HCs (Table 1). Patients with HIV-SN (14 males, 0 females), CIPN (19 males, 1 female), or CIDP (22 males, 12 females) were predominantly male compared to those having DPN (12 males, 17 females), for whom there was a slightly higher prevalence among females. Glycated hemoglobin (HbA1c) was within normal range, indicating no diabetes for HCs (38 ± 3.8 mmol/mol) or for patients with HIV-SN (within normal range), CIPN (34.8 ± 5.8 mmol/mol), or CIDP (35.9 ± 6.8). HbA1c was high for patients with DPN indicating diabetes (72.4 ± 19.3 mmol/mol). Disease duration was longer for HIV-SN (25.3 ± 4.7 years), CIDP (12 ± 8.6 years), and DPN (42 ± 14.4 years) but shorter for CIPN (<6 months), as expected.

Table.

Demographic and CCM Parameter Results

	Mean	SD	HC (n = 20)	HIV-SN (n = 14)	CIPN (n = 34)	CIDP (n = 29)
Age (y)
HC	54.5	11.2
HIV-SN	57.7	7.8	0.99
CIPN	63.8	10.0	0.06	—
CIDP	59.2	14.8	0.52	—	—
DPN	58.5	12.0	0.71	—	—	—
CNFD (fibers/mm2)
HC	39.4	6.7
HIV-SN	26.7	4.2	<0.0001
CIPN	29.5	5.5	0.004	0.99
CIDP	25.1	7.7	<0.0001	0.79	0.55
DPN	17.4	9.9	<0.0001	0.002	<0.0001	0.001
CNBD (branches/mm2)
HC	95.9	40.2
HIV-SN	74.0	22.0	0.6
CIPN	73.6	43.3	0.61	0.99
CIDP	80.3	40.2	0.77	0.99	0.99
DPN	45.7	31.7	<0.0001	0.18	0.22	0.003
CNFL (mm/mm2)
HC	27.3	4.7
HIV-SN	22.2	3.4	0.16
CIPN	21.3	5.0	0.05	0.99
CIDP	19.4	6.0	0.0001	0.8	0.98
DPN	14.6	8.2	<0.0001	0.002	0.0133	0.02
CNFrD (unitless)
HC	1.51	0.02
HIV-SN	1.51	0.02	0.99
CIPN	1.50	0.03	0.99	0.99
CIDP	1.46	0.04	0.0009	0.02	0.12
DPN	1.40	0.08	<0.0001	<0.0001	<0.0001	0.001
ACNFrD (unitless)
HC	18.0	3.0
HIV-SN	14.7	2.2	0.07
CIPN	14.2	3.1	0.03	0.79
CIDP	13.4	4.0	0.0007	0.67	0.78
DPN	10.2	5.3	<0.0001	0.005	0.01	0.01

P < 0.05 was considered significant.

Corneal Confocal Microscopy

The detailed results are presented in Table 1 and Figure 3. CNFD was significantly lower in patients with HIV-SN (P < 0.0001), CIPN (P = 0.004), CIDP (P < 0.0001), or DPN (P < 0.0001) compared to HCs and was lower in DPN compared to HIV-SN (P = 0.002), CIPN (P < 0.0001), or CIDP (P = 0.001) (Fig. 3A). CNBD was significantly lower only in DPN compared to HCs (P < 0.0001) or CIDP (P = 0.003) (Fig. 3B). CNFL was significantly lower in patients with CIPN (P = 0.05), CIDP (P = 0.0001), or DPN (P < 0.0001) compared to HCs and was lower in DPN compared to HIV-SN (P = 0.002), CIPN (P = 0.01), or CIDP (P = 0.02) (Fig. 3C). CNFrD was significantly lower in patients with CIDP compared to HCs (P = 0.0009) or HIV-SN (P = 0.02) and in DPN compared to HCs (P < 0.0001), HIV-SN (P < 0.0001), CIPN (P < 0.0001), or CIDP (P = 0.001) (Fig. 3D). ACNFrD was lower in patients with CIPN (P = 0.03), CIDP (P = 0.0007), or DPN (P < 0.0001) compared to HCs and differed in DPN compared to HIV-SN (P = 0.005), CIPN (P = 0.01), or CIDP (P = 0.01) (Fig. 3E).

Figure 3.

Box and whisker plots of corneal nerve morphology parameters in controls and neuropathy groups. The box represents the mean (solid line) with 25th and 75th percentiles; whiskers represent 5th and 95th percentiles for CNFD (A), CNBD (B), CNFL (C), CNFrD (D), and the CNFL/CNFrD ratio (ACNFrD) (E), with black circles representing outliers in the sample. One-way analysis of variance was used to compare between groups.

Discussion

To the best of our knowledge, this is the first study to compare established measures of corneal nerve loss and CNFrD²⁶ in peripheral neuropathies of differing etiology. There are two main findings in this study. First, patients with DPN or CIDP showed more severe corneal nerve loss compared to patients with HIV-SN or CIPN, who had milder albeit significant loss. Our second and most important finding is that the remaining corneal nerves may have a distinct morphological pattern, as assessed by CNFrD in DPN and CIDP compared to HIV-SN and CIPN. Furthermore, even after adjustment for the overall severity of reduction in CNFL, ACNFrD differed significantly in patients with DPN compared to CIDP, HIV-SN, and CIPN.

Peripheral neuropathy occurs in the setting of many common and rare neurologic disorders. It affects approximately 26% to 58% of patients with diabetes,^34,35 up to 50% of people living with HIV,³⁶ and 68% of patients with cancer after chemotherapy.³⁷ CIDP affects 5 out of 100,000 people.^38,39 The corneal subbasal nerve plexus is a dense network of unmyelinated C-fibers, which are anatomically similar to intra-epidermal nerve fibers,^40,41 the gold standard markers for assessing small fiber integrity. Unlike intra-epidermal innervation, which requires an invasive skin biopsy, corneal subbasal innervation can be quantified rapidly and non-invasively using in vivo CCM. We have previously shown that corneal nerve loss occurs in DPN,^2–9 CIDP,¹⁶ HIV-SN,¹⁷ and CIPN.¹⁸ More recently, we and others have also shown corneal nerve loss in patients with Parkinson's disease,⁴² multiple sclerosis,^43–45 and dementia.⁴⁶ These findings highlight the potential of CCM as a surrogate measure of neurodegeneration but also raise concerns regarding the clinical utility of this technique due to limited specificity.

Fractal dimension analysis is a measure of tissue structural complexity, and in brain magnetic resonance imaging scans^28–31 it has been used to identify distinct tissue geometric patterns in different neurodegenerative diseases. In ophthalmology research, fractal dimension analysis has been used to study alterations in the retinal vascular network in diabetic retinopathy⁴⁷ and maculopathy⁴⁸ and to detect glaucoma in digital images.^49,50 A recent study has used spatial analysis to show increased clustering of corneal nerve loss in patients with early diabetic neuropathy⁵¹ and we have shown altered CNFrD in patients with DPN.²⁶ In this context, our findings suggest that in addition to demonstrating corneal nerve loss, there may be specific imaging traits of remaining nerve fibers associated with the underlying etiology.

Strengths of this study are the detailed clinical phenotyping of neuropathy and the homogeneous imaging protocol used between centers. Given that ex vivo⁴¹ and in vivo CCM studies^52,53 have shown that nerves in the inferior cornea are highly tortuous with a whorl-like appearance compared to almost straight nerves in the central cornea, all measurements were performed from the central cornea using a standardized protocol. A main limitation is the use of an automated image analysis algorithm to estimate CNFrD. Because CNFrD depends on accurate nerve segmentation, it may have been underestimated compared to the true fractal dimension.⁴ Another limitation is the relatively small sample size, particularly for people living with HIV and CIPN, which may have skewed the results toward milder neuropathy. Finally, as only images from the central region were quantified for CNFrD, we cannot exclude the possibility of image selection bias. Assessment of wide-field CCM images may provide additional value for the estimation of CNFrD. Nevertheless, we show that fractal dimension analysis could complement the existing toolbox of corneal nerve analysis metrics and help to differentiate peripheral neuropathies, especially CIDP and DPN. We believe the additional quantification of CNFrD enhances the clinical utility of CCM by enhancing the specificity of this technique. Future studies should assess whether corneal nerve fractal dimension can differentiate central from peripheral neurodegenerative disorders.