. 2022 Jan 7;16(2):295–320. doi: 10.1177/19322968211035375

Table 4.

Corneal Confocal Microscopy (CCM)

Author(s) & study aim	Study population & design	Study outcomes & implications
Alam et al.⁷² - Compare diagnostic capability of CCM against IENFD & QST in patients with DSPN	N = 88 n = 30 T1D, -DSPN M = 38.8 ± 12.5 n = 31 T1D, +DSPN M = 53.3 ± 11.9 n = 27 HCs M = 41.0 ± 14.9 Cross-sectional	• AUC for CNFD was 0.81 with an optimal cutoff of 25.0 no/mm², sensitivity of 77% & specificity of 79% for diagnosis of DSPN • AUC for CNBD was 0.67 with an optimal cutoff of 36.5 no/mm², sensitivity of 58% & specificity of 79% for diagnosis of DSPN • AUC for CNFL was 0.74 with an optimal cutoff of 16.8 mm/mm², sensitivity of 61% & specificity of 80% for diagnosis of DSPN • AUC for IENFD was 0.73 with an optimal cutoff of 4.5 fibers/mm, sensitivity of 61% & specificity of 86% for diagnosis of DSPN • CCM is a non-invasive method, with respectable performance, to identify small nerve fiber pathology & diagnose DSPN
Al-Fahdawi et al.⁹² -Propose a robust, fast & fully automatic nerve segmentation & morphometric parameter quantification system for CCM images	N = 20, 498 images n = 12 +DM M = 58 ± 10 yrs/age n = 4 -DPN n = 5 mild DPN n = 3 moderate DPN n = 8 HCs M = 54 ± 7 yrs/age Cross-sectional	• For manually & automatically traced nerves, average tortuosity was 8.27 ± 7.18 & 6.7 ± 5.60, respectively, for controls; 20.11 ± 19.04 & 13.9 ± 12.79, respectively, for -DPN group; 37.52 ± 36.41 & 29.32 ± 28.36, respectively, for mild DPN group; & 40.45 ± 39.30 & 51.76 ± 50.64, respectively, for moderate DPN group • For manually & automatically traced nerves, average nerve length was 60.92 mm & 61.22 mm, respectively, for controls; 58.49 mm & 56.87 mm, respectively, for -DPN group; 57.08 mm & 56.87 mm, respectively, for mild DPN group; & 57.08 mm & 56.63 mm, respectively, for moderate DPN group • Across manually & automatically traced nerve methods, comparable results found for average nerve length values although average tortuosity values appear less consistent • Both manually & automatically traced nerve methods reveal increasing nerve tortuosity & decreased nerve length according to severity of DPN
Al Rashah et al.⁹³ -Assess repeatability CNM rate measurement in individuals with or without neuropathy	N = 14 M = 51.9 ± 13.8 yrs/age n = 5 +DM, +DN n = 5 +DM, -DN n = 4 HCs Longitudinal (3-week FU)	• Within & between observer repeatability of CNM within 4 different zones & CNM rate of the vertical section of the wide-field montage was outstanding with ICCs of 0.99 & 0.99, respectively • Repeatability of CNM rate of the vertical section of the wide-field montage within observers, when using semi-automated & fully automated image montaging software, was also outstanding with an ICC of 0.96 • With a laser-scanning CCM, CNM rate measurement shows impressive repeatability for within & between observers & when using manual, semi-automated, & fully automated image montaging
Azmi, et al.⁷³ -Assess whether baseline and follow-up measures of neuropathy, particularly small-fiber neuropathy, relate to changes in glucose tolerance over 3 yrs	N = 47 n = 30 IGT M = 60 ± 2.1 yrs/age n = 17 Controls M = 62.3 ± 1.8 yrs/age Longitudinal (3-year FU)	• FU: 10 IGT participants developed T2D & had significantly lower CNFD, CNBD, & CNFL at baseline compared to controls; IGT participants who developed T2D also had a further significant reduction in CNFL, IENFD, & MDL • FU: 15 participants had no change in IGT status & 5 participants returned to NGT with no significant baseline abnormality on CCM or IENFD • FU: IGT participants (n=15) showed a significant decrease in IENFD but no change in CCM • FU: Participants returning to NGT (n=5) showed a significant increase in CNFD, CNBD, & CNFL, but a significant decrease in IENFD • CCM may be an early marker of small-fiber neuropathy & allow for risk stratification of individuals with IGT likely to progress to T2D
Batawi et al.⁹⁴ - Study SCNP parameters by IVCCM using a new software technology	N = 29 M = 46 ± 20 yrs/age n = 22 +DM Cross-sectional	• Inter-operator reproducibility & intra-operator reproducibility ICCs, respectively, were: (1) NFLD 0.97 & 0.97; (2) NFL 0.97 & 0.97; (3) NF 0.90 & 0.93; (4) NT 0.97 & 0.96; (5) NBi 0.83 & 0.87; (6) NBr 0.81 & 0.87; (7) BD 0.92 & 0.95; (8) NBe 0.95 & 0.97; (9) NFT 0.73 & 0.88 • Semi-automated corneal nerve analysis showed excellent precision or reproducibility for evaluation of SCNP parameters with CCM
Brines et al.⁹⁵ -Using an automated methodology, compare sensitivity & specificity of NFD, NBD, NFL, & NFA for discriminating participants with neuropathy from NCs	N = 129 n = 21 -DN M = 37.1 ± 16.5 yrs/age n = 21 mild DN M = 55.9 ± 11.0 yrs/age n = 19 moderate DN M = 59.0 ±11.3 yrs/age n = 20 severe DN M = 57.0 ± 14.6 yrs/age n = 48 NCs M = 46.2 ± 16.9 yrs/age Cross-sectional	• In participants with no or mild DN, relative to NCs, respectively, AUROC curves for NFD, NFB, NFL, NFA WxL, & NFA FIJI (≥0.70 – <0.85 range) were fair to good • In participants with moderate to severe DN, relative to NCs, respectively, AUROC curves for NFD, NBD, NFL, & NFA WxL (>0.80 – <1.0 range) were good to excellent • With DM participants collapsed into 1 group, cutoff points for discrimination of participants having DN from NCs were calculated from ROC curves; DM group cut point for NFD was 23.4 fibers/mm² with a sensitivity of 90% & specificity of 69%; DM group cut point for NBD was 23.4 branches/mm² with a sensitivity of 86% & specificity of 71%; DM group cut point for NFL was 12.3 mm/mm² with a sensitivity of 96% & specificity of 68%; DM group cut point for NFA FIJI was 19,128 μm²/mm² with a sensitivity of 94% & specificity of 59% • Overall, automated methodology revealed good to excellent diagnostic performance in detecting moderate to severe DN & impressive ability to identify true positives for DN in DM participants
Chen, et al.⁹⁶ -Determine the diagnostic performance of CCM (manual & automated methods) & IENFD for DSPN	N = 89 n = 17 T1D, +DSPN M = 59 ± 11 yrs/age n = 46 T1D, -DSPN M = 44 ± 13 yrs/age n = 26 CG M = 44 ± 15 yrs/age Cross-sectional	• AUCs: CNFD manual 0.82 & automated 0.80; CNFD manual S/S = 76% at EERP; CNFD manual S/S = 82%/71% at threshold of 24.0 (M ± 2 SDs); CNFD automated S/S = 70% at EERP; CNFD automated S/S = 60%/83% at threshold of 15.5 (M ± 2 SDs) • AUCs: CNFL manual 0.70 & automated 0.77; CNFL manual S/S = 71% at EERP; CNFL manual S/S = 59%/74% at threshold of 16.5 (M ± 2 SDs); CNFL automated S/S = 70% at EERP; CNFL automated S/S = 59%/80% at threshold of 10.5 (M ± 2 SDs) • AUCs: CNBD manual 0.59 & automated 0.70; CNBD manual S/S = 53% at EERP; CNBD manual S/S = 17%/96% at threshold of 15.0 (M ± 2 SDs); CNBD automated S/S = 59% at EERP; CNBD automated S/S = 29%/98% at threshold of 4.0 (M ± 2 SDs) • AUC of IENFD = 0.66; IENFD S/S = 65% at EERP; IENFD S/S = 53%/76% at threshold of 3.3 (M ± 2 SDs) • Overall, performance between CCM (manual & automated) & IENFD comparable
Chen, et al.⁹⁷ -Evaluate an automated software tool for nerve-fiber detection & quantification in CCM images, combining sensitive nerve-fiber detection with morphological descriptors	N = 176; 888 images n = 63 T1D, +DSPN n = 29 T1D, -DSPN n = 84 Controls Cross-sectional	• AUCs of ACNFD, ACNFL, & ACNBD were 0.76, 0.76, 0.68, respectively; ACNFD S/S = 65%, ACNFL S/S = 62%, & ACNBD S/S = 58% at EEP for discriminating +DSPN & -DSPN • AUCs of MCFND, MCNFL, & ACNBD were 0.79, 0.71, & 0.61, respectively; MCNFD S/S = 72%, MCNFL S/S = 66% & MCNBD S/S = 59% at EEP for discriminating +DSPN & -DSPN • AUCs of combined CCM automated & manual features were 0.74 & 0.78, respectively; CCM automated S/S = 71%; CCM manual S/S = 68% for discriminating +DSPN & -DSPN • Performance of automated quantification compared to manual quantification of corneal nerves in CCM images is relatively equivalent • Automated quantification provides a sensitive tool for identification of DSPN while improving speed & repeatability
Chen, et al.⁹⁸ -Evaluate the performance of previously established CCM parameters to a novel automated measure of corneal nerve complexity (ACNFrD)	N = 176 n = 84 AMCs M = 46 ± 15 yrs/age n = 29 T1D, +DSPN M = 63 ± 12 yrs/age n = 63 T1D, -DSPN M = 44 ± 15 yrs/age Cross-sectional	• AUROC curve for ACNFD was 0.77, ACNFL was 0.74, ACNBD was 0.69, & ACNFrD was 0.74; ACNFD S/S = 65% at EER; S/S = 63%/79% at threshold of 15.1(M ± 2 SDs); ACNFL S/S = 62% at EER; S/S = 62%/83% at threshold of 10.2 (M ± 2 SDs); ACNBD S/S = 58% at EER; S/S = 24%/98% at threshold of 3.3 (M ± 2 SDs); ACNFrD S/S = 65% at EER; S/S = 61%/78% at threshold of 1.45 (M ± 2 SDs) • AUROC curves were 0.79 for MCNFD, 0.71 for MCNFL, & 0.61 for MCNBD; MCNFD S/S = 72% at EER; S/S = 79%/71% at threshold of 23.8 (M ± 2 SDs); MCNFL S/S = 65% at EER; S/S = 55%/86% at threshold of 14.9 (M ± 2 SDs); MCNBD S/S = 59% at EER; S/S = 17%/96% at threshold of 13.8 (M ± 2 SDs) • ACNFrD performance comparable to automated & manual CNFD, CNBD, & CNFL in diagnosing patients with +DSPN or -DSPN • Automated & manual methods show good equivalency in accuracy
Dehghani et al.⁹⁹ - Determine alterations in CSNP over 4 yrs using IVCCM	N = 108 T1D, -PN (baseline) M = 43.9 ± 15.7 age/yrs (baseline) Longitudinal	• At 4-yr FU, mean CNFD (no./mm², 19.6 ± 6.9) was marginally significantly increased compared to baseline (18.3 ± 7.1) • At 4-yr FU, mean CNBD (no./mm², 29.1 ± 19.6) was significantly increased relative to baseline (24.2 ± 17.4) • At 4-yr FU, mean CNFL (mm/mm², 16.3 ± 3.7) did not significantly change from baseline (16.0 ± 3.8) • IVCCM may be useful for monitoring subclinical alterations in CSNP in T1D
Dell’Omo, et al.¹⁰⁰ -Measure the thickness & length of corneal nerves & peri-papillary RNFL thickness in patients recently diagnosed with T2D	N = 44 n = 22 new onset T2D M = 50.6 ± 6.74 age/yrs n = 22 HCs M = 50.8 ± 4.26 age/yrs Cross-sectional	• Evaluated by CCM, CNT & CNL were significantly lower in participants with new onset T2D than controls • Using SD-OCT, RNFL thickness in temporal quadrant was significantly lower in patients with T2D relative to controls • ICC values for intra-observer and inter-observer repeatability for CNT were 0.76 & 0.78 & for CNL 0.71 & 0.74, respectively • Reductions observed in CNL, CNT, & RNFL thickness suggest that CCM & SD-OCT may detect early markers of neuropathy in patients recently diagnosed with T2D • ICC values revealed good reliability
Dhage, et al.⁷⁵ -Assess the longitudinal utility of different measures of neuropathy in patients with diabetes	N = 38 n = 19 +DM M = 52.5 ± 14.7 yrs/age (baseline) n = 19 HCs M = 47.4 ± 14.2 yrs/age (baseline) Longitudinal cohort study (MFU = 6.5 yrs)	• At baseline, CNFD, CNBD & CNFL significantly reduced in DM participants vs. controls • Compared to baseline, CNFD, CNBD & CNFL significantly decreased in DM participants at FU • At baseline, NSP & NDS scores significantly higher & IENFD significantly lower in DM participants vs. controls • Compared to baseline, NSP & NDS scores significantly increased & IENFD significantly decreased in DM participants at FU • Change in CNFD, CNBD, & CNFL significantly correlated with change in IENFD • CCM, a rapid, non-invasive & reproducible ophthalmic imaging technique to objectively quantify small-fiber damage in DN, was found to longitudinally identify progressive nerve damage
Edwards et al.¹⁰¹ -Demonstrate DN development & progression in T1D individuals	N = 38, T1D, -DN (baseline) Longitudinal cohort study (4-yr FU)	• CNFL identified 7 participants who had unfavorable corneal nerve changes from baseline to 4-yr FU • CNFL, relative to PNCV, CST, WST, VPT, NDS, & monofilament testing, consistently performed better • CNFL revealed the earliest, most sustained, & highest proportion of abnormal parameters indicative of PN development
Ferdousi et al.¹⁰² - Identify longitudinal CNM changes in CC & IW relative to other DN measures	N = 36 n = 19 age-matched HCs M = 49.47 ± 13.25 yrs/age n = 30 +DM M = 54.08 ± 15.86 yrs/age (baseline) n = 21 T1D n = 9 T2D Longitudinal cohort MFU = 3.6 ± 1.3	• In participants with DM, CNBD (mm/mm²) & CNFL (mm/mm²) were significantly reduced from baseline (57.72 ± 30.08; 21.77 ± 5.19, respectively) to FU (44.04 ± 23.69; 15.65 ± 4.7, respectively) • In participants with DM, IWL (mm/mm²) & ANFL (CNFL + IWL/2; mm/mm²) were also significantly reduced from baseline (24.69 ± 8.67; 23.26 ± 5.53, respectively) to FU (14.23 ± 6.13; 15.09 ± 4.48, respectively • Rate of annual decline in CNFL, IWL, & ANFL was significantly higher in patients with DM compared to controls • ICC showed good consistency between the changes per year in CNFL & IWL in participants with DM (ICC = 0.78; 95% confidence interval, 0.56–0.88)
Ferdousi, et al.⁷⁸ -Compare the utility of quantifying corneal nerve loss at the inferior whorl & central cornea to QST & NCS in the diagnosis & assessment of DPN severity	N = 143 n = 93 +DM n = 51 -DPN M = 57.68 ± 1.6 yrs/age n = 47 Mild DPN M = 60.16 ± 1.7 yrs/age n = 45 Moderate to severe DPN M = 64.1 ± 1.48 yrs/age	• ROC curve & Youden index used to define optimum cutoff points for CNFD, CNBD, CNFL, & IWL; CNFD AUC 0.71, sensitivity 58%, & specificity 83%; CNBD AUC 0.70, sensitivity 69%, & specificity 65%; CNFL AUC 0.68, sensitivity 64%, & specificity 67%; IWL AUC 0.72, sensitivity 70%, & specificity 65% • CNFD & CNBD were significantly lower in patients with -DPN (26.61 ± 1.05 & 64.07 ± 4.39, respectively) & mild DPN (24.47 ± 1.09 & 58.49 ± 4.76, respectively) compared to controls (33.71 ± 1.3 & 81.52 ± 5.54, respectively); CNFD & CNBD were significantly lower in patients with moderate to severe DPN (22.4 ± 1.14 & 45.60 ± 4.5, respectively) relative to those with -DPN & controls
	n = 30 Controls M = 54.51 ± 2.3 yrs/age Cross-sectional	• CNFL & IWL were significantly lower in patients with mild (20.84 ± 1.00 & 22.28 ± 1.31, respectively) & moderate to severe (19.27 ± 1.04 & 19.03 ± 1.36, respectively) DPN compared to controls (25.07 ± 1.27 & 31.69 ± 1.66, respectively); CNFL & IWL were significantly lower in patients with moderate to severe DPN relative to those with -DPN (23.31 ± 0.96 & 24.90 ± 1.26, respectively) • CCM identifies early & progressive corneal nerve loss at the inferior whorl & CC
Ferdousi, et al.¹⁰³ -Assess the diagnostic utility of CCM for DPN & risk factors for corneal nerve loss	N = 490 n = 149 T1D n = 269 T2D n = 72 HCs Cross-sectional	• CNFD, CNBD, & CNFL were significantly reduced in participants with T1D vs. those with T2D • ROC curve analysis for diagnosis DPN showed a very good AUC for CNFD at 0.81 with an optimal cutoff point of 29.40/mm²; sensitivity was 73.5% & specificity was 74.4% • ROC curve analysis for diagnosis DPN showed a reasonable AUC for CNBD at 0.74 with an optimal cutoff point of 64.58/mm²; sensitivity was 66.7% & specificity was 66.7% • ROC curve analysis for diagnosis DPN showed a reasonable AUC for CNFL at 0.73 with an optimal cutoff point of 24.00 mm/mm²; sensitivity was 66.7% & specificity was 66.4% • CCM identified more severe corneal nerve loss in T1D patients relative to those with T2D & shows very reasonable diagnostic accuracy for DPN
Guimarães et al.¹⁰⁴ -Describe & validate an automatic approach to nerve tracing	N = 30 n = 24 pathologic group n = 10 +DM n = 6 HCs Cross-sectional	• Automatic nerve tracing, following a proposed algorithm, was assessed with respect to manual grading of CNT (high, mid, & low tortuosity classes) with a significant Spearman’s rank correlation coefficient of 0.95 achieved • With setting 2 thresholds to distinguish between the 3 tortuosity classes, correct classification (93.3%) was yielded for the automatic compared to the manual approach • Nerve tracing results reveal the automatic method performed well although larger studies are needed to confirm results
Kalteniece et al.¹⁰⁵ -Assess the effect of a standardized protocol for image selection & number of images required for adequate quantification of CN pathology using IVCCM	N = 35 obese &/or +DM M = 49.97 ± 12.47 yrs/age Longitudinal	• In terms of inter-observer variability, ICC values for CNFD, CNBD, & CNFL were significant at 0.93, 0.96, & 0.95, respectively • With respect to intra-observer variability, ICC values for CNFD, CNBD, & CNFL were significant at 0.95, 0.97, & 0.97, respectively • For sample size variability, ICC values for CNFD, CNBD, & CNFL were significant at 0.94, 0.95, & 0.96, respectively • Bland-Altman plots showed excellent agreement for all parameters • 6 images were found to be adequate for the fully automated analysis • Implementing a standardized protocol to select IVCCM images results in high intra- & inter-observer reproducibility for all corneal nerve parameters
Lewis et al.106 -Determine reference distribution of annual CNFL change, prevalence of abnormal change in diabetes, & associated variables	N = 794 n = 399 T1D M = 39.9 ± 18.7 yrs/age n= 191 T2D M = 60.4 ± 8.2 n = 204 Controls M = 37.9 ± 19.8 yrs/age Secondary analysis of longitudinal observational study	• Participants with T1D (median FU of 3 visits over 4.4 yrs) had a mean annual change in CNFL (mm/mm²) of -0.8%; T1D participants with RCNFL had a significant annual change in CNFL (-14.67 ± 11.46%) relative to T1D participants without RCNFL (2.58 ± 9.93%) • Participants with T2D (median FU of 3 visits over 5.3 yrs) had a mean annual change in CNFL of -0.2%; T2D participants with RCNFL had a significant annual change in CNFL (-11.49 ± 6.35%) compared to T2D participants without RCNFL (2.47 ± 7.33%) • RCNFL prevalence was 17% overall & similar by DM type (16.0% T1D, 19.4% T2D) • RNCFL was significantly more frequent in those with baseline DSP (47%) vs. those without baseline DSP (30%) • RCNFL may identify patients at high risk for DSP development & progression
Li et al.¹⁰⁷ -Examine & compare fully-automated & manually measured corneal nerve fiber parameters in T2D patients with & without DPN	N = 152 n = 128 T2D n = 49 -DPN M = 67.12 ± 6.01 yrs/age n = 79 +DPN M = 70.15 ± 7.34 yrs/age n = 24 HCs M = 68.63 ± 5.19 yrs/age Cross sectional	• CNFL_M & CNBD_M in +DPN group were significantly lower than -DPN & HC groups • Likewise, CNFL_FA & CNBD_FA in +DPN group were significantly reduced relative to -DPN & HC groups • CNFD_FA, but not CNFD_M, was significantly reduced in +DPN group vs. HC group • Significant, positive correlations between manual & fully automated CNFL, CNFD, & CNBD were observed • Fully automated method slightly underestimated corneal nerve fiber parameters. • A progressive decrease in manual & fully automated CNFL, CNBD & CNFD accompanied the occurrence of DPN • Fully automated corneal nerve fiber parameter quantification may be a fast, objective way to detect DPN
Lovblom et al.¹⁰⁸ -Determine the predictive validity of a baseline IVCCM measure in identifying future DSP onset in patients with T1D	N = 65 T1D, -DSP M = 34 ± 15 yrs/age Longitudinal (mean 3.5 years FU)	• At FU, 54 (83%) remained without DSP & 11 (17%) developed DSP • New-onset cases of DSP had lower baseline CNFL & CNBD but higher baseline CNFT • CNFL: AUC was 0.78; optimal operating threshold of 14.9 mm/mm² with 82% sensitivity, 69% specificity, & 35% PPV • CNFT: AUC was 0.72; optimal operating threshold of 15.4 (tortuosity coefficient) with 73% sensitivity, 72% specificity, & 35% PPV • CNBD: AUC was 0.71; optimal operating threshold of 36.1 branches/mm² with 82% sensitivity, 50% specificity, & 25% PPV • CNFL may have applicability in identifying high-risk patients for DSP
Ostrovski et al.¹⁰⁹ -Determine inter- & intra-observer reproducibility of a novel automated analysis program vs. manual analysis	N = 46 n = 26 T1D M = 42.8 ± 16.9 yrs/age n = 20 controls M = 41.4 ± 17.3 yrs/age	• Inter-observer ICCs for CNFL_M, CNFL_SA, & CNFL_FA were 0.73, 0.75 & 0.78, respectively, with no significant differences for 3-way comparisons; intra-observer ICCs were 0.72, 0.74, & 0.84, respectively, with CNFL_FA reproducibility significantly higher than that of CNFL_M & CNFL_SA • Inter-observer & intra-observer ICCs for CNFD_M, CNFD_SA, CNFD_FA, CNBD_M, CNBD_SA, & CNBD_FA were substantially lower compared to those for CNFL • Fully automated analysis preserves CNFL reproducibility despite an apparent measurement bias (underestimation) relative to the manual strategy of image analysis
Perkins et al.¹¹⁰ - Establish concurrent validity & diagnostic thresholds in a large cohort of participants with & without DSP	N = 998 n = 516 T1D M = 42 ± 19 yrs/age n = 482 T2D M = 62 ± 10 yrs/age Cross-sectional, pooled multi-national consortium study	• In participants with T1D, CNFL_A had an AUC of 0.77 & optimal threshold of 12.5 mm/mm² (73% sensitivity & 69% specificity) • In participants with T2D, CNFL_A had an AUC of 0.68 & optimal threshold of 12.3 mm/mm² (69% sensitivity & 63% specificity) • In participants with T1D, AUC for CNBD_A & CNFD_A were 0.73 & 0.71, respectively • In participants with T2D, AUC for CNBD_A & CNFD_A were 0.66 & 0.52, respectively • In the total cohort, CNFL_A had an AUC of 0.71 & optimal threshold of 12.3 mm/mm² (67% sensitivity & 66% specificity); AUC of CNFL_M (0.70) vs. CNFL_A was marginally, yet significantly lower although its optimal threshold value of 16.3 mm/mm² had similar operating characteristics • A lower CNFL_A threshold value of <8.6 mm/mm² to rule in DSP & upper CNFL_A threshold value of 15.3 mm/mm² to rule out DSP was associated with 88% specificity & 88% sensitivity • In the largest cohort to date, diagnostic validity & common diagnostic thresholds for CNFL in T1D & T2D established
Petropoulos et al.¹¹¹ - Compare ability of CNFD, CNFL, & IWL alone & in combination for diagnosis of DPN	N = 68 n = 53 +DM (T1D & T2D) n = 25 +DPN M = 60.1 ± 10.2 yrs/age n = 28 –DPN M = 42.4 ± 14.7 yrs/age n = 15 AMCs Cross-sectional	• For diagnosis of DPN, ROC curve analysis showed CNFL = 21.9 mm/mm² had an AUC of 0.75, sensitivity of 76%, & specificity of 65%; CNFD = 28.4 fibers/mm² had an AUC of 0.74, sensitivity of 68%, & specificity of 61%; IWL = 20.0 mm/mm² had an AUC of 0.70, sensitivity of 68%, & specificity of 67% • Combination of CNFL & IWL achieved an AUC of 0.75, sensitivity of 80%, & specificity of 71% for DPN • For inter & intra-observer agreement for IWL estimation, no significant difference between 2 separate observers (17.8 ± 8.5 vs. 17.7 ± 9.1 mm/mm²) & no significant difference (17.8 ± 8.5 vs. 17.2 ± 8.0 mm/mm²) when same observer assessed & reassessed IW center images from 1 dataset on 2 separate occasions • CNFD, CNFL, & IWL have comparable ability to diagnose DPN; combining IWL & CNFD may improve CCM diagnostic performance • Inter- & intra-observer agreement for IWL estimation was excellent
Pritchard et al.⁸¹ -Determine if deficits in CNFL assessed using CCM can predict future onset of DPN	N = 90 T1D, -DPN (baseline) 4-yr FU n = 16 +DPN M = 51 ± 14 yrs/age (baseline) n = 64 -DPN M = 42 ± 16 yrs/age (baseline) Longitudinal	• DPN developed in 16 participants (18%) after 4 yrs • Baseline CNFL (mm/mm²) predicted incident DPN at 4-yr FU • Participants who developed DPN at 4-yr FU had significantly lower baseline CNFL (14.0 ± 4.1) relative to participants (16.2 ± 3.5) who did not develop DPN • For CNFL, AUROC was 0.6 with a threshold cutoff of 14.1 mm/mm²; sensitivity was 63% & specificity was 74% to predict DPN • CCM may predict DPN in T1D
Pritchard et al.¹¹² -Compare SNF damage in central cornea & whorl area in participants with DPN & examine accuracy of evaluating these 2 sites for DPN diagnosis	N = 187 n = 107 T1D M = 48.3 ± 15.1 yrs/age n = 25 +DPN n = 82 -DPN n = 80 Controls M = 37.0 ± 17.8 yrs/age Cross-sectional	• Participants with & without DPN had significantly lower CNFL_center compared to controls (14.2 ± 3.5 & 16.7 ± 3.5 mm/mm² vs. 19.3 ± 3.0 mm/mm², respectively) • Participants with & without DPN had significantly reduced CNFL_whorl relative to controls (15.4 ± 4.4 & 18.2 ± 3.9 mm/mm² vs. 22.1 ± 3.9 mm/mm², respectively) • For CNFL_center, AUC was 0.76; Youden Index cutoff point of <17.9 with sensitivity of 90% & specificity of 50% for detecting DPN • For CNFL_whorl, AUC was 0.77; Youden Index cutoff point of <18.6 with sensitivity of 80% & specificity of 60% for detecting DPN • SNF pathology comparable at corneal central & whorl anatomical sites • Quantification of CNFL from the corneal center is as accurate as CNFL quantification of whorl area for diagnosis of DPN
Scarpa et al.¹¹³ - Investigate whether CNN can be successfully used for corneal nerve multiple-image analysis	N = 100 participants, 600 confocal images M = 53 ± 13 yrs/age n = 50 +DM (T1D & T2D), +DN n = 50 AMHCs Cross-sectional	• Cross-validation used to evaluate a proposed algorithm for binary classification (healthy or pathological) of 100 participants (CNN training on 80 subjects & evaluation on the other 20, repeated 5 times) with 97% mean accuracy for CNN correct classification • Also, with a final classification derived for each participant (properly classified if both right & left eyes correctly classified), CNN revealed validity with a mean accuracy of 96% • With outstanding classification accuracy, the CNN is a highly promising, fully automated method of corneal confocal image analysis with strong potential to yield improved results obtained from traditional methods
Scarr et al.¹¹⁴ -Evaluate agreement between manual & automated analysis protocols in controls & T1D & T2D participants with & without DSP	N = 456 M = 53 ± 18 yrs/age n = 139 T1D n = 249 T2D n = 68 Controls Cross-sectional	• For the study population, mean CNFL_M (15.1 ± 4.9 mm/mm²) was greater than mean CNFL_A (10.5 ± 3.7 mm/mm²) although values were highly correlated; absolute mean difference between CNFL_A & CNFL_M for the study population was -4.6 ± 2.6 mm/mm² & percentage difference was -29 ± 17%, representing underestimation bias by CNFL_A • A similar pattern of correlations & underestimation bias was observed for CNFL_M vs. CNFL_A for T1D, T2D, & control groups • In participants with T1D, percentage difference between CNFL_A & CNFL_M for those with DSP was -27 ± 27%, & for those without DSP the percentage difference was -32 ± 13% although non-significantly • In participants with T2D, percentage difference between CNFL_A & CNFL_M for those with DSP was -28 ± 16%, & for those without DSP was -27 ± 16% although non-significantly • Weighted kappa statistic for agreement between tertiles of CNFL_A & CNFL_M was 0.62, indicating moderate to substantial agreement • CNFL_A & CNFL_M were significantly lower in T1D participants with DSP relative to those without DSP • CNFL_A & CNFL_M were not significantly lower in T2D participants with DSP compared to those without DSP • Although CNFL_A underestimated CNFL_M, its bias was non-differential between participant groups & its relationship with DSP status was preserved • Determination of diagnostic thresholds specific to CNFL_A requires further investigation
Schaldemose et al.¹¹⁵ -Compare a new sampling method & AAC with established methods of corneal nerve quantification in patients with & without DSPN & HCs	N = 88 n = 62 T1D n = 17 +DSPN M = 59 ± 11 yrs/age n = 45 -DSPN M = 44 ± 13 yrs/age n = 26 HCs M = 44 ± 15 yrs/age Cross-sectional	• Using randomized & area adjusted method, CNFD_M & CNFL_M were significantly reduced in +DSPN group compared to both HC & -DSPN groups; CNFL_M values were larger in -DSPN group relative to HCs • Using a randomized & area adjusted method, CNFD_A, CNFL_A &, CNBD_A were reduced significantly in +DSPN & -DSPN groups compared to HCs & lowest in the +DSPN group • Randomized sampling & adjusted area method with automated analysis showed that, among +DSPN participants, CNFD_A (no./mm²; 17.3 ± 12) had a higher mean than standard automated procedures (13.5 ± 9.1) with a significant difference of 28.1%; CNFL_A (mm/mm²; 12.3 ± 6.8) had a higher mean than standard automated procedures (8.8 ± 4.7) with a significant difference of 39.8%%; CNBD_A (no./mm²; 19.1 ± 14) had a higher mean than standard automated procedures (15.4 ± 12) with a significant difference of 24.0% • Randomized sampling & adjusted area method with automated analysis showed that, among -DSPN participants, CNFD_A (no./mm²; 28.2 ± 9.3) had a higher mean than standard automated procedures (22.6 ± 7.3) with a significant difference of 24.8%; CNFL_A (mm/mm²; 17.0 ± 4.2) had a higher mean than standard automated procedures (13.4 ± 3.3) with a significant difference of 26.9%%; CNBD_A (no./mm²; 31.1 ± 18) had a higher mean than standard automated procedures (26.2 ± 15) with a significant difference of 18.7% • Interobserver reliability testing revealed no significant difference in means for CNFL_M, CNFD_M, & CNBD_M between the investigator & a blinded second observer • Randomized sampling method & area-dependent analysis objectively shows a reduction in corneal nerve parameters in diabetic patients with & without DSPN while providing unbiased corneal nerve quantification
Tummanapalli et al.¹¹⁶ - Determine utility of CCM & tear neuromediator analysis in diagnosis of DPN	N = 70 n = 38 T1D n = 19 +DPN M = 44 ± 12 age/yrs n = 19 -DPN M = 37 ± 10 age/yrs n = 32 T2D n = 16 +DPN M = 58 ± 6 age/yrs n = 16 -DPN M = 52 ± 12 age/yrs Prospective, cross- sectional	• For DPN diagnosis in T1D, IWL had the best diagnostic performance (AUC 0.91, optimal diagnostic threshold of ≤12.93 mm/mm², 89% sensitivity, & 83% specificity), followed by CNFrD (AUC 0.89; diagnostic threshold ≤1.50, 89% sensitivity, & 72% specificity), CNFD (AUC 0.88, diagnostic threshold ≤22.46 no./mm², 83% sensitivity, & 83% specificity), IWNFrD (AUC 0.88, diagnostic threshold ≤1.46, 83% sensitivity, & 89% specificity), CNFL (AUC 0.87, diagnostic threshold ≤14.24 mm/mm², 89% sensitivity, & 78% specificity), CNBD (AUC 0.76, diagnostic threshold ≤26.17 no./mm², 72% sensitivity, & 72% specificity), & CTBD (AUC 0.75, diagnostic threshold ≤40.40 no./mm², 72% sensitivity, & 72% specificity) • For DPN diagnosis in T2D, CNFL had the best diagnostic performance (AUC 0.81, optimal diagnostic threshold ≤13.64 mm/mm², 81% sensitivity, & 81% specificity), followed by CNFrD (AUC 0.78, diagnostic threshold ≤1.48, 75% sensitivity, & 63% specificity), CNBD (AUC 0.76, diagnostic threshold ≤29.17 no./mm², 81% sensitivity, & 75% specificity) CNFD (AUC 0.76, diagnostic threshold ≤22.90 no./mm², 75% sensitivity, & 69% specificity), CTBD (AUC 0.76, diagnostic threshold ≤43.10 no./mm², 81% sensitivity, & 81% specificity), & IWNFrD (AUC 0.73, diagnostic threshold ≤1.47, 69% sensitivity, & 62% specificity) • In T1D & T2D, different corneal nerve parameters were identified, suggesting different disease processes across conditions
Wang et al.¹¹⁷ -Explore the application of CCM in DPN & other T2D chronic complications	N = 220 n = 100 T2D +DPN M= 56.2 ± 12.4 age/yrs n = 72 -DPN M = 54.9 ± 11.1 age/yrs n = 37 T2D +DN n = 89 T2D +DR n = 48 HCs M = 51.9 ± 14.9 yrs/age Cross-sectional	• In participants with T2D, CNFL & CNBD were significantly lower than HCs, & similar in +DPN group compared to -DPN group • Using CCM to identify DPN, AUC for CNFD was 0.67 & optimal cutoff was 24.68 no./mm² (78% sensitivity, 53% specificity); AUC for CNFL was 0.70 & optimal cutoff was 15.32 mm/mm² (80% sensitivity, 60% specificity); AUC for CNBD was 0.68 & optimal cutoff was 39.0 no./mm² (85% sensitivity, 47% specificity) • This study provides more support for clinical use of CCM to diagnose DPN
Williams et al.¹¹⁸ -Validate a DLA for corneal nerve segmentation in CCM images & compare to the widely used, validated automated image analysis software, ACCMetrics	N = 3865 confocal images of corneal SBP obtained from +DM individuals & HCs n = 120 confocal images from Italy n = 1578 confocal images from China n = 2137 confocal images from UK	• DLA, employing CNN with data augmentation, trained using a high-end graphics processor unit on 1698 images for automated quantification of the corneal SBP for DN diagnosis • DLA further tested on 2137 images for external validation with DLA ICCs superior to those for ACCMetrics for total CNFL (0.93 vs. 0.83), mean length per segment (0.66 vs. 0.33), number of branch points (0.89 vs 0.57), number of tail points (0.62 vs. 0.26), number of nerve segments (0.88 vs. 0.50) & fractals (0.93 vs 0.76) • DLA achieved an AUC of 0.83, specificity of 0.87, & sensitivity of 0.68 for the classification of participants without (n = 90) or with (n = 132) neuropathy • Results reveal DLA provides rapid & excellent localization performance for quantification of corneal nerve biomarkers • DLA model has potential for adoption into clinical screening programs for DN

Abbreviations: AAC, adjusted area calculation; ACNBD, automated corneal nerve branch density; ACNFD, automated corneal nerve fiber density; ACNFL, automated corneal nerve fiber length; ACNFrD, automated corneal nerve fiber fractal dimension; AMC, age-matched controls; AMHCs, age-matched healthy controls; ANFL, average nerve fiber length; AUC, area under the curve; AUROC, area under the receiver operator characteristic; BD, beading density; CC, central cornea; CCM, corneal confocal microscopy; CG, control group; CN, corneal nerve; CNBD, corneal nerve branch density; CNBD_A, corneal nerve branch density with automated analysis; CNBD_FA, corneal nerve branch density with fully automated analysis; CNBD_M, corneal nerve branch density with manual analysis; CNBD_SA, corneal nerve branch density with semi-automated analysis; CNFD, corneal nerve fiber density; CNFD_A, corneal nerve fiber density with automated analysis; CNFD_FA, corneal nerve fiber density with fully automated analysis; CNFD_M, corneal nerve fiber density with manual analysis; CNFD_SA, corneal nerve fiber density with semi-automated analysis; CNFL, corneal nerve fiber length; CNFL_A, corneal nerve fiber length with automated analysis; CNFL_center, corneal nerve fiber length at the central cornea; CNFL_FA, corneal nerve fiber length with fully automated analysis; CNFL_M, corneal nerve fiber length with manual analysis; CNFL_SA, corneal nerve fiber length with semi-automated analysis; CNFLwhorl, corneal nerve fiber length at the whorl; CNFrD, corneal nerve fractal dimension; CNFT, corneal nerve fiber tortuosity; CNL, corneal nerve length; CNM, corneal nerve migration; CNN, convolutional neural network; CNT, corneal nerve thickness or corneal nerve tortuosity; CSNP, corneal sub-basal nerve plexus; CTBD, corneal total branch density; DLA, deep learning algorithm; DM, diabetes mellitus; DN, diabetes neuropathy; DPN, diabetes peripheral neuropathy; DSP, diabetic sensorimotor polyneuropathy or distal symmetric polyneuropathy; DSPN, diabetes symmetrical peripheral neuropathy or diabetic sensorimotor polyneuropathy; EEP, equal error point; EER, equal error rate; EERP, equal error rate point; FU, follow-up; HCs, healthy controls; ICC, interclass correlation coefficient; IENFD, intraepidermal nerve fiber density; IGT, impaired glucose tolerance; IVCCM, in vivo corneal confocal microscopy; IWL, inferior whorl length; IWNFrD, inferior whorl nerve fractal dimension; M, mean; MCNBD, manual corneal nerve branch density; MCNFD, manual corneal nerve fiber density; MCNFL, manual corneal nerve fiber length; MDL, mean dendric length; MFU, mean follow-up; mm, millimeter; mm/mm², millimeter/square millimeter; mm², millimeters squared; NBD, nerve branch density; NBe, number of beadings; NBi, number of bifurcations; NBr, number of branches; NCs, normal controls; NDS, Neuropathy Disability Score; NFA FIJI, nerve fiber area determined by total number of pixels within the nerve plexus; NFA W×L, nerve fiber area determined by length times average width; NFD, nerve fiber density; NFL, nerve fiber length; NFLD, nerve fiber length density; NFT, nerve fiber tortuosity; NGT, normal glucose tolerance; no., number; no./mm², number/square millimeter; NSP, Neuropathy Symptom Profile,; NT, number of trunks; PN, peripheral neuropathy; PNCV, peroneal nerve conduction velocity; PPV, positive predictive value; QST, quantitative sensory testing; RCNFL, rapid corneal nerve fiber loss; RNFL, retinal nerve fiber layer; ROC, receiver operator characteristic; S/S, sensitivity/specificity; SBP, sub-basal nerve plexus; SCNP, sub-basal corneal nerve plexus; SD, standard deviation; SD-OCT, spectral domain optical coherence tomography; SNF, small nerve fiber; T1D, type 1 diabetes; T2D, type 2 diabetes; VPT, vibration perception threshold; vs., versus; WST, warm sensation threshold; yr, year; yrs, years; μm², square micrometer.