Снимки нервной системы в системной нейропатии: мини-обзор экспериментальных и точных методов лечения.

Core Tip: Corneal confocal microscopy provides a fast, non-invasive quantitative assessment of corneal small-nerve morphology and has shown reproducible group-level abnormality in diabetic peripheral neuropathy and a few other systemic neuropathies. Nevertheless, population-based distinctions do not mean a reliable diagnosis in a single patient. Research shows that when it comes to diabetic neuropathy, the best studied condition, the variability in diagnostic performance can be attributed to the studied population, the reference standard employed, the sampling protocol used, the analysis chosen and the threshold selected. Also, there is considerable overlap and variation between those affected and not affected.

INTRODUCTION

The cornea is densely innervated and provides an accessible window into peripheral small-fiber integrity. In individuals with impaired glucose tolerance, those who later developed type 2 diabetes showed reductions in corneal nerve fiber density, branch density, and length, whereas those who reverted to normal glucose tolerance showed improvement[ 1 ]. In established diabetes, corneal confocal microscopy (CCM) abnormalities generally become more pronounced with increasing neuropathy severity and may precede abnormalities on conventional clinical or neurophysiological testing[ 2 - 5 ]. Reduced corneal nerve parameters have also been reported in Parkinson’s disease, particularly in association with autonomic involvement[ 6 - 8 ], and in early relapsing-remitting multiple sclerosis[ 9 ]. These findings support investigation of CCM as a noninvasive marker of systemic small-fiber involvement.

Neurotoxic chemotherapy can also alter corneal nerve morphology. An initial clinical report described CCM abnormalities in a patient with chemotherapy-induced sensory neuropathy[ 10 ]. A subsequent prospective study found longitudinal worsening of corneal nerve parameters alongside neuropathy measures after chemotherapy, supporting the potential of CCM as a noninvasive structural marker of chemotherapy-induced peripheral neuropathy[ 11 ]. These observations remain exploratory and require validation in larger disease- and treatment-specific cohorts.

CCM may also be useful for longitudinal assessment of nerve repair. In type 1 diabetes, simultaneous pancreas-kidney transplantation was followed by increases in corneal nerve fiber density and length within 6 months[ 12 ]. A subsequent systematic review and meta-analysis reported improvement in corneal nerve parameters after several pharmacological and surgical interventions, supporting CCM as a potential research endpoint for nerve regeneration[ 13 ]. Automated and artificial intelligence (AI)-based approaches may further improve the speed and reproducibility of quantification, and early classifiers have shown potential for diabetic neuropathy classification[ 14 - 17 ].

Overall, these data support further evaluation of CCM as a rapid, noninvasive, objective, and reproducible adjunctive biomarker of small-fiber injury, particularly in diabetic peripheral neuropathy (DPN); however, the available evidence does not establish disease-specific patient-level diagnostic validity. The growing literature underscores its clinical relevance, although broader translation into routine practice still depends on standardized protocols for image acquisition, sampling, selection, and analysis, as well as the availability of age-adjusted normative reference values[ 18 - 22 ]. Consequently, this review is located predominantly within experimental and translational medicine and appraises CCM as a quantitative biomarker of systemic small-fiber injury. Corneal nerve morphology may offer a in vivo window on peripheral nerve involvement in systemic and neurologic disease because the cornea contains a dense and readily accessible network of small sensory nerve fibers ( Figure 1 ). Primary corneal neuropathy disorders which affect the cornea or ocular surface as the primary site of disease are beyond the scope of this review, and shall only be considered when relevant to anatomic context, methodologic interpretation or risk of confounding. The aim of this minireview is to discuss on the state of the evidence surrounding CCM for the detection, diagnosis, risk stratification, longitudinal assessment, response-to-treatment evaluation and so forth of systemic small-fiber neuropathies, while alluding to the methodological pitfalls that ought to be resolved prior to implementation in practice.

Existing reviews have largely emphasized disease-specific diagnostic performance, technical aspects of CCM, or selected applications such as nerve regeneration. The present minireview instead organizes the evidence around five clinically distinct questions: Early detection, patient-level diagnostic evaluation, prognostic stratification, longitudinal monitoring, and treatment-response assessment. Its incremental contribution is not a new diagnostic meta-analysis, but an integrated translational framework that combines quantitative patient-level performance, methodological standardization, ocular confounding, and AI-related bias to define the boundary between group-level association and individual clinical utility. In this framework, translational medicine refers to the evaluation of CCM as a candidate biomarker moving from research toward clinical validation, whereas precision medicine refers specifically to the potential use of validated patient-level CCM information for stratification or decision support. CCM is therefore considered complementary when it adds structural small-fiber information to established testing, unsuitable as a substitute when conventional diagnostic confirmation remains indicated, and inappropriate for stand-alone decision-making when thresholds are unvalidated, ocular confounding is unresolved, or disease-specific evidence is insufficient.

SEARCH STRATEGY

The objective of this minireview was not to conduct a systematic review or meta-analysis, but rather to conduct a structured narrative review. The primary bibliographic database was PubMed/MEDLINE because the goal was to develop a clinically focused synthesis of biomedical evidence across heterogeneous disease settings and study designs, rather than an exhaustive systematic evidence search. Two searches were conducted, with the final search performed on June 1, 2026. The following set of terms was combined for the search: (“corneal confocal microscopy” OR “in vivo corneal confocal microscopy” OR “corneal nerve imaging” OR “corneal nerve fiber”) AND (“small fiber neuropathy” OR “small-fiber neuropathy” OR “diabetic neuropathy” OR “diabetic peripheral neuropathy” OR “Parkinson disease” OR “Parkinson’s disease” OR “multiple sclerosis” OR “chemotherapy-induced peripheral neuropathy” OR “peripheral neuropathy”) AND (“diagnosis” OR “early detection” OR “prognosis” OR “longitudinal monitoring” OR “treatment response” OR “nerve regeneration” OR “artificial intelligence” OR “automated analysis” OR “precision medicine”). To mitigate the risk of failing to detect important studies, we screened reference lists of eligible reviews, systematic reviews, meta-analyses, and key original studies. Two authors screened and selected the studies and extracted relevant data independently, with disagreements resolved by consensus. Since the review was narrative rather than systematic, PRISMA reporting and a formal PRISMA flow diagram were not applied; the structured search is reported to enhance transparency and reproducibility of study identification only and does not imply systematic-review completeness. However, the single-database strategy remains a limitation and a potential source of selection bias.

Eligible publications were original human clinical or translational study, systematic review, meta-analysis or a clinically relevant narrative review evaluating quantitative corneal nerve imaging relevant to systemic small-fiber or peripheral neuropathy. Studies that reported any of the following: Corneal nerve fiber density, corneal branch density, corneal nerve fiber length, corneal nerve fiber tortuosity, inferior whorl measurements, diagnostic performance, longitudinal change and treatment-related nerve regeneration, were included in the review. We included studies of the imaging methodology, normative values, reproducibility, automated analysis, or artificial intelligence that were directly relevant to interpreting or clinically translating measurements. We excluded reports that dealt exclusively with large-fiber neuropathy and primary corneal disease or an ocular surface disorder without relevant systemic neuropathy. Only those case reports and small case series that provide historical importance or mechanistic or hypothesis-generating information were considered. We also considered whether studies reported or controlled clinically relevant ocular factors capable of independently modifying corneal nerve measurements, particularly ocular surface disease, contact lens wear, previous ocular surgery, and primary corneal disease. Incomplete reporting or control of these variables was considered a potential source of confounding when interpreting associations between CCM findings and systemic neuropathy.

We screened titles and abstracts for relevance and then full texts of possibly eligible papers. A narrative synthesis was conducted across five a priori clinical domains: Early detection, diagnostic evaluation, prognostic stratification, longitudinal monitoring, and treatment-response assessment. Due to considerable disparity in study populations, imaging protocols, outcome definitions, reference standards, and analysis methods, quantitative pooling was not undertaken.

Due to the variability of the included designs, no single formal risk-of-bias instrument was applied to all studies. Instead, evidence was evaluated according to prespecified characteristics relevant to clinical credibility. The features considered were study design, whether data were collected prospectively or retrospectively, sample size, single-center or multicenter setting, external validation, appropriateness of the clinical/neurophysiological/pathological reference standard, and control of relevant ocular confounders. Also considered was consistency with independent studies. Greater interpretive weight was given to systematic reviews, meta-analyses, multicenter studies, and prospective longitudinal cohorts. By contrast, small cross-sectional studies, case series, and case reports were viewed as exploratory or hypothesis-generating; these sources were not used alone to support patient-level diagnostic claims.

The standardized terminology and measurement units used to describe corneal nerve morphology are summarized in Table 1 . To enhance comparability across studies, we paid particular attention to the imaging device and acquisition mode used, anatomical region examined, number and area of images analyzed, image-selection strategy, image quality criteria, analysis software, and use of manual, semiautomated, or fully automated quantification.

Anatomy of corneal nerves and its pathophysiological connection to systemic neuropathy

Corneal sensory innervation arises predominantly from the ophthalmic division of the trigeminal nerve. Stromal nerve trunks enter radially from the limbus, lose their myelin sheaths and perineurium, branch anteriorly, penetrate Bowman’s layer, and form the dense subbasal nerve plexus between Bowman’s layer and the basal epithelium. Most quantitative CCM studies of systemic neuropathy focus on this plexus because its relatively planar architecture is well suited to reproducible imaging and morphometry. Corneal sensory nerves consist mainly of small Aδ and C fibers and contribute to sensation, blink reflexes, tear secretion, epithelial turnover, and wound healing[ 23 - 27 ]. Through release of neuropeptides, growth factors, and other mediators, these nerves also support epithelial trophism and ocular-surface homeostasis[ 28 ]. This anatomical accessibility and small-fiber composition provide the rationale for evaluating corneal nerve morphology as a marker of peripheral nerve integrity[ 25 , 29 , 30 ]. In this review, the cornea is therefore considered an accessible tissue for assessing systemic small-fiber integrity rather than the primary site of a corneal neuropathic disorder.

Chronic hyperglycemia, in fact, activates several alternative glucose metabolic pathways: For example, the metabolism of polyols, protein kinase C, hexosamines, and advanced glycation products, which determine alterations and redox imbalances with mitochondrial activity and accumulation of toxic metabolites[ 31 - 34 ]. These processes are accompanied by microvascular impairment, reduced endoneuria perfusion, and ischemic stress, all of which further compromise axonal and Schwann cell integrity[ 32 - 34 ]. At the same time, low-grade intraneural inflammation, marked by macrophage activation and increased cytokine signaling, appears to sustain ongoing degeneration, particularly in small nerve fibers[ 34 , 35 ]. In type 2 diabetes, dyslipidemia, obesity, and insulin resistance may intensify these mechanisms, reinforcing the idea that diabetic neuropathy arises from converging rather than isolated pathogenic pathways[ 33 , 34 ].

Neurodegenerative disorders share several mechanisms relevant to axonal injury, including mitochondrial dysfunction, oxidative stress, altered proteostasis, and impaired axonal transport[ 36 , 37 ]. Because neuronal survival depends on sustained bioenergetics and long-distance intracellular trafficking, distal axons and small sensory fibers may be particularly vulnerable. CCM studies have reported corneal nerve loss in several peripheral neuropathies and selected neurodegenerative diseases, supporting investigation of the subbasal plexus as an accessible marker of broader neuroaxonal injury[ 29 , 30 ].

Accumulating evidence indicates that corneal nerve abnormalities can be detected early in dysglycemia and diabetic neuropathy, in some cases before neuropathy becomes evident according to conventional clinical criteria[ 1 , 5 ]. In individuals with impaired glucose tolerance, reduced corneal nerve fiber density and length have been documented, particularly in those who later progressed to type 2 diabetes, despite largely comparable nerve conduction findings[ 1 ]. Together with meta-analytic evidence showing corneal nerve loss even in diabetic patients without clinically established neuropathy, these observations suggest that corneal imaging can reveal subclinical small-fiber damage that may be missed by traditional diagnostic approaches[ 4 , 5 , 18 ]. Longitudinal studies further indicate that corneal nerve parameters are dynamic markers of small-fiber injury rather than static descriptors. In subjects with impaired glucose tolerance, lower baseline corneal nerve fiber density, branch density, and length were observed in those who later developed type 2 diabetes, whereas significant improvements in these measures were observed in individuals who reverted to normal glucose tolerance[ 1 ]. In people with diabetic neuropathy, loss of corneal nerves associated with central and inferior whorl is reported more[ 38 ].

Progress in corneal confocal microscopy and imaging technologies

Over the past two decades, in vivo CCM has evolved from a predominantly research technique into a rapid, noninvasive method for high-resolution assessment of the cornea[ 18 , 39 , 40 ]. Modern laser-scanning systems, particularly the Heidelberg Retina Tomograph-Rostock Corneal Module, enable quantitative analysis of subbasal nerve morphology. The principal CCM metrics and reporting conventions used in this review are defined in Table 1 and are not repeated here. Manual and automated approaches can provide reproducible quantification, with automated methods reducing analysis time and observer dependence[ 14 , 15 , 41 ]. Wide-field composite imaging expands spatial sampling, while emerging techniques such as micro-optical coherence tomography may further improve visualization of corneal nerves[ 42 - 44 ].

Technical challenges and unresolved issues in corneal nerve image analysis

The measurement of CCM is still sensitive to acquisition and sampling. The areas covered by single frames of the spatially heterogeneous subbasal plexus are small, and measurements may vary across central, paracentral, and inferior-whorl areas. Many other factors like eye movement, focus, depth of imaging, illumination, contrast, and compression of tissue can also affect image quality. Longitudinal studies also face a challenge in determining biological change due to the difficulty of relocating the same microscopic region, making biological change difficult to distinguish from spatial sampling variability[ 21 , 39 , 42 , 45 ]. Strategies for image selection also differ, as in representative, random, consecutive nonoverlapping, central, inferior-whorl, and wide-field sampling. Observer bias can be introduced by manual selection, while overlapping frames can duplicate nerve segments. Wide-field mosaics allow sampling error to be reduced, although extra registration, de-duplication, and analysis steps are required[ 21 , 22 , 39 , 42 , 45 ].

The software packages also differ in their analytical definitions, like how to distinguish between nerve trunks and branches, how to identify crossings and branch points, how to deal with fibers that are cut off at the image edge, and how to determine tortuosity. Low-contrast or discontinuous nerves can easily be missed by such algorithms. Reflective cells and imaging artifacts can also be classified as nerve fibers. Thus, measurements obtained through different software, preprocessing steps, segmentation thresholds, and parameter definitions may not be interchangeable[ 15 , 39 , 45 , 46 ].

Clinical applications and evidence domains in systemic neuropathies

To clarify, available evidence is discussed in relation to five key clinical applications, namely, early detection of neuropathy, diagnostic work-up, prognostic stratification, disease course monitoring, and assessment of response to therapy. To distinguish the maturity of evidence across these applications, we use three descriptive categories rather than formal Grading of Recommendations Assessment, Development and Evaluation ratings: Relatively established, emerging, and exploratory. Evidence is relatively established for group-level detection and complementary diagnostic assessment in diabetic neuropathy; emerging for longitudinal monitoring and treatment-response assessment; and exploratory for prognostic prediction and most disease-specific applications outside diabetes. These categories reflect study design, sample size, external validation, reference standard, and independent replication and should not be interpreted as formal certainty-of-evidence grades.

Early detection of neuropathy

Using traditional clinical or neurophysiological criteria, corneal nerve abnormalities may be detectable before neuropathy. In people with impaired glucose tolerance, reductions in the density of corneal nerve fibers, corneal nerve branch density, and length have been reported, particularly in those who later developed type 2 diabetes[ 1 ]. Evidence from meta-analyses and cohorts also shows that corneal nerve loss may occur in patients with diabetes without clinically evident neuropathy[ 5 , 47 - 49 ]. Previous studies reported similar early abnormalities in relapsing-remitting multiple sclerosis, but the evidence in that setting is more limited[ 9 ]. These findings suggest that CCM may detect subclinical small-fiber involvement, although the abnormalities observed are not disease-specific.

Diagnostic evaluation

The diagnostic evidence for diabetic neuropathy is strongest, but separation at the group level does not necessarily translate into accuracy at the individual level. A systematic review and meta-analysis of 38 studies with almost 4000 participants showed that corneal nerve fiber density, branch density, fiber length, and inferior whorl length were decreased in a graded manner across controls, diabetes without clinically established neuropathy, and established DPN. For corneal nerve fiber length, the pooled mean difference between established neuropathy and diabetes without neuropathy was -3.08 mm/mm 2 (95%CI: -3.58 to -2.58)[ 5 ]. A meta-analysis of 52 studies in non-diabetic neuropathies similarly found lower corneal nerve fiber length in clinical and subclinical neuropathy groups, but disease heterogeneity precluded a common diagnostic threshold[ 30 ].

Patient-level performance is more variable. In the study with 998 participants from the multinational consortium, automated corneal nerve fiber length yielded an area under the curve (AUC) of 0.77 in type 1 diabetes and an AUC of 0.68 in type 2 diabetes; in the overall cohort, a threshold of 12.3 mm/mm 2 yielded an AUC of 0.71, with a sensitivity and specificity of 67% and 66%, respectively, with a considerable intermediate range remaining unclassified[ 48 ]. Performance also depended on the comparator: One cohort reported an AUC of 0.84 when abnormal peroneal motor nerve amplitude was used as the comparator, but approximately 0.67-0.69 when warm-detection thresholds were used[ 14 ].

In more recent or mild type 2 diabetic polyneuropathy, the sensitivity of corneal nerve fiber length was 14.4%, but specificity was high at 95.7% (AUC 0.55), and it did not improve diagnostic performance over other small-fiber measures[ 47 ]. Wide-field CCM likewise exhibited limited correlation with concurrent skin biopsy or functional tests[ 50 ]. In a study of 680 unselected neurological patients, sensitivity and specificity of CCM were 44% and 75%, respectively, with an AUC of 0.63, while skin-biopsy specificity was 99% and its AUC was 0.74[ 51 ]. CCM must be viewed as a complementary structural measure that serves as neither a stand-alone diagnostic test nor a direct substitute for established small-fiber assessments.

Prognostic stratification

Prognostic evidence remains limited. In a 3-year cohort of 30 participants with impaired glucose tolerance and 17 controls, 10 progressed to type 2 diabetes, 15 remained with impaired glucose tolerance, and five reverted to normal glucose tolerance. Progressors had lower baseline corneal nerve fiber density, branch density, and length, whereas those reverting to normal glucose tolerance showed improvement[ 1 ]. However, only 10 participants progressed, no externally validated prediction model was developed, and the outcome was progression to diabetes rather than incident or worsening neuropathy. These findings are therefore hypothesis-generating for metabolic trajectory rather than validation of CCM as a prognostic biomarker.

Evidence in Parkinson’s disease is predominantly cross-sectional. A study of 98 patients and 26 controls found lower corneal nerve fiber density, branch density, total branch density, and fiber length, without consistent associations with disease duration, motor severity, stage, subtype, cognition, or quality of life[ 6 ]. Another study found greater corneal nerve loss with increasing autonomic involvement, with AUCs of 0.872 and 0.915 for discrimination between current autonomic phenotypes[ 8 ]. These findings describe cross-sectional phenotyping rather than prediction of future progression.

Longitudinal disease monitoring

CCM is well suited to serial assessment because it is noninvasive and repeatable[ 21 , 22 ]. Longitudinal changes have been reported in impaired glucose tolerance[ 1 ] and after neurotoxic chemotherapy[ 11 ]. However, serial CCM findings should be interpreted alongside symptoms, neurological examination, neurophysiology, skin biopsy, and other small-fiber measures rather than as an isolated marker[ 30 , 50 , 51 ].

Assessment of therapeutic response

CCM may also detect morphological nerve regeneration after intervention. In type 1 diabetes, corneal nerve fiber density and length increased within 6 months of simultaneous pancreas-kidney transplantation[ 12 ], and a subsequent systematic review and meta-analysis reported improvement in corneal nerve parameters after several pharmacological and surgical interventions[ 13 ]. These changes support CCM as a potential research endpoint, but short-term morphological improvement is not a validated surrogate for long-term neurological recovery or clinical benefit.

The available evidence therefore varies considerably according to the clinical use. The strongest support is for early detection and assessment of diabetic neuropathy with complementary diagnostic tests. Prognostic stratification, longitudinal monitoring of disease, and assessing treatment response require further evaluation in prospective studies. Table 2 summarizes representative quantitative evidence and patient-level diagnostic performance across the principal clinical settings.

Precision medicine and AI in corneal nerve evaluation

The integration of corneal nerve imaging into precision medicine is supported by its potential to provide objective, noninvasive, and repeatable biomarkers of small-fiber injury and repair, particularly in diabetic neuropathy and in longitudinal studies of treatment response[ 18 , 19 , 48 ]. Precision medicine depends on reliable patient-level biomarkers, and CCM-derived measures can quantify corneal nerve morphology with good repeatability when image acquisition and analysis are performed using standardized protocols[ 21 , 22 ]. AI-based methods may further strengthen this approach by enabling faster and potentially more consistent image classification. Even so, broader clinical adoption still depends on external validation, technical standardization, and a clearer definition of how these tools should complement established diagnostic tests rather than simply replicate them[ 41 , 47 , 48 ].

AI has increasingly been applied to CCM analysis to reduce reliance on labor-intensive manual assessment and improve consistency[ 15 , 16 , 41 ]. Automated and deep-learning approaches have shown good agreement with expert annotation[ 14 - 16 ]. In a cohort of 222 participants, including 90 without and 132 with neuropathy according to the Toronto criteria, a deep-learning classifier achieved an AUC of 0.83, sensitivity of 0.68, and specificity of 0.87[ 16 ]. A subsequent three-class study tested the algorithm in 40 participants, including 15 healthy volunteers, 13 participants without peripheral neuropathy, and 12 with peripheral neuropathy. For the peripheral-neuropathy class, the model achieved a recall of 0.83 (95%CI: 0.58-1.00), precision of 1.00 (95%CI: 1.00-1.00), and F1 score of 0.91 (95%CI: 0.74-1.00)[ 17 ]. These results are encouraging but derive from selected datasets and require larger prospective external validation. Current AI evidence, so far, does not yet support autonomous clinical deployment, despite these results. Most CCM AI studies to date are based on selected datasets with limited diversity in centers, devices, acquisition protocols, and patient populations. Not all published literature lacks external validation; for instance, Williams et al [ 16 ] included an external image-validation dataset. However, prospective clinical validation across independent centers, different CCM devices, operators, acquisition settings, and demographic groups remains limited[ 16 , 17 , 41 , 46 ]. It is important to make this distinction because good performance within an external image dataset does not necessarily establish transportability across real-world clinical domains.

Another crucial methodological risk is data leakage. A common scenario in CCM is obtaining multiple frames from the very same participant. Hence, splitting by random image-level assignment can lead to placing highly correlated images from the same subject into both training and test datasets, which may inflate apparent performance. Training, validation, and test sets should therefore be separated at the participant level, and preferably at the center level when considering true external generalizability. Class imbalance, selection of images based on quality, preprocessing procedures, annotation variability, and device differences might introduce additional bias or domain shift[ 41 , 46 ].

Interpretability stays unresolved too. A classifier’s high accuracy does not mean that the algorithm has learned biologically meaningful neuropathological features. Deep-learning systems can use acquisition artifacts, illumination, focus, image quality, and other non-causal image characteristics that correlate with study labels. Attribution or saliency maps may help visualize model behavior, but they do not demonstrate biological or clinical validity on their own. Systems used in the clinic should therefore provide transparent image-quality control, interpretable morphometric outputs when possible, patient-level predictions, calibration and uncertainty estimates, and external validation rather than an isolated categorical diagnosis[ 41 , 46 ]. AI must show added value beyond traditional CCM morphometry and existing neuropathy assessments before claims of clinical implementation can be justified.

Overall, AI has the potential to reinforce the role of corneal nerve imaging within precision medicine, but wider clinical implementation will still require larger external validation studies, standardization, and effective regulatory translation[ 17 , 41 ]. Figure 2 summarizes the main strengths, current limitations, and future priorities of corneal nerve imaging.

DISCUSSION

There is current evidence that supports CCM as a non-invasive structural marker of small-fiber injury, with the strongest data in diabetic neuropathy[ 1 , 5 , 18 , 30 , 48 ]. However, correlations with skin biopsy, functional measures, and neuropathy severity are inconsistent, especially in cases of milder or recently diagnosed type 2 diabetes[ 47 , 50 , 51 ]. Thus, CCM must be considered complementary rather than a stand-alone surrogate of the status of the peripheral nerve. This review considers the corneal subbasal plexus to be an accessible surrogate tissue for systemic small-fiber injury, differentiating it from primary corneal neuropathic diseases in which the cornea is the target of disease[ 18 , 40 ].

Because of the fact that CCM directly measures the cornea, local ocular conditions can also modify the same morphometric endpoints that are interpreted as markers of systemic small-fiber injury. Dry eye disease (DED) is relevant. A recent systematic review and meta-analysis found central corneal nerve length and nerve density to be significantly reduced in DED compared to healthy controls (pooled mean differences approximately -4.0 mm/mm 2 and -7.2 nerves/mm 2 , respectively[ 52 ]).

The emergence of a CCM phenotype overlapping with systemic neuropathy can be caused by ocular surface disease that is unidentified or poorly described. Primary corneal diseases such as keratoconus or ectasia, infectious or inflammatory keratitis, corneal dystrophies, neurotrophic keratopathy, and previous corneal injury might also influence corneal innervation[ 24 , 25 ]. Another important modifier is prior eye surgery. Corneal refractive surgeries transect or disrupt corneal nerves directly and are associated with procedure-dependent patterns of reinnervation[ 53 ]; cataract surgery has also been associated with changes in the subbasal nerve plexus[ 54 ]. It is necessary to document contact lens wear too. Evidence on conventional soft and rigid contact lenses is varied. However, nerve distribution and corneal sensitivity show more pronounced alterations with orthokeratology and contact-lens-related ocular surface disease[ 55 ].

It is possible that these factors have research and clinical implications. In studies of CCM as a marker of systemic neuropathy, ocular history and ocular-surface assessment should be prespecified. The study should document clinically relevant DED, contact lens type and use, previous ocular surgery and time since surgery, topical medication, previous corneal trauma or infection, and primary corneal disease. Patients with active primary corneal disease or who have had recent procedures that are likely to substantially alter corneal innervation should generally be excluded from diagnostic-accuracy analyses or analyzed separately. In the case of a more common condition such as DED, contact lens wear, or remote ocular surgery, exclusion may not always be appropriate. In such circumstances, prespecified stratification, adjustment, or sensitivity analyses may be more useful as they are better at preserving external validity. In CCM neuropathy studies, there is no universally validated washout interval for these factors; thus, arbitrary time-based exclusions should be avoided. In the clinical context, abnormal CCM values should not be attributed to systemic neuropathy in the absence of an ocular history and a slit-lamp/ocular-surface examination. When ocular exclusion criteria are poorly defined, the association of corneal nerve abnormalities with systemic disease should be viewed more cautiously. As a methodological limitation, this narrative synthesis relied on a single bibliographic database and did not apply design-specific formal risk-of-bias instruments, which could lead to selection or appraisal bias.

Table 3 summarizes the main strengths, current limitations, and future priorities of corneal nerve imaging.

Research supports CCM as a marker of DPN. According to[ 1 , 5 , 48 ], the evidence base is clearest for DPN. It is supported by a number of systematic reviews, meta-analyses, multicenter diagnostic studies, and longitudinal cohorts. Patient-level performance and agreement with established small-fiber measures still vary between populations in this setting[ 47 , 50 , 51 ], and there is currently no evidence of CCM-guided management independently improving clinical outcomes. Evidence in Parkinson’s disease, multiple sclerosis, and chemotherapy-induced peripheral neuropathy is mostly observational and exploratory[ 6 - 11 ]. CCM is best viewed as a complementary biomarker. Its routine implementation would require standardized acquisition and analysis, validated thresholds, appropriate reference standards, and clinically meaningful longitudinal endpoints.

A highly relevant application of CCM may be longitudinal and interventional studies. CCM parameters might serve as responsive research endpoints, owing to corneal nerve regeneration following pancreas transplantation and other interventions[ 12 , 13 ]. Nevertheless, morphological improvement must not be considered a validated surrogate for neurological recovery; future studies should further couple CCM with symptoms and findings from neurological examination, skin biopsy, and neurophysiology, along with patient-reported outcomes[ 13 , 47 , 51 ].

CCM adoption in standard clinical practice also depends on practical factors including dedicated instrumentation, operator training, acquisition time, image-quality control, automated analysis, interoperability, and reimbursement. These barriers may be particularly relevant in settings with limited resources where dedicated CCM hardware, trained operators, standardized analysis software, and reimbursement pathways may be restricted[ 18 , 39 , 40 ]. The use of AI may lessen manual burdens and improve repeatability, but improved classification performance alone is not enough. In fact, algorithms should be validated externally across devices, populations, and disease contexts. In addition, there should be transparency in reporting, interpretability, and regulatory assessment[ 16 , 17 , 41 ]. Future CCM-AI research should also follow contemporary principles of prediction-model reporting and validation, such as transparent partitioning of participant-level data, model-performance and calibration reporting, external validation, and explicit assessments of risk of bias and applicability, consistent with TRIPOD+AI and PROBAST+AI[ 56 , 57 ].

Corneal nerve imaging will most likely achieve clinical value when it is incorporated into clearly defined multidisciplinary pathways in which it answers a specific diagnostic or monitoring question and is interpreted in conjunction with established assessments of small-fiber disease[ 18 , 30 ]. Operationally, CCM is best positioned after clinical evaluation as a noninvasive structural adjunct: Concordant findings may strengthen phenotyping or longitudinal monitoring, whereas discordant or diagnostically consequential findings should be interpreted alongside nerve-conduction studies and/or quantitative sensory testing and, when confirmation of small-fiber neuropathy is required, skin biopsy[ 18 , 30 , 47 , 51 ].

CONCLUSION

Corneal nerve imaging is a promising non-invasive method to evaluate small-fiber injury in systemic neuropathies. Evidence has been strongest in diabetic neuropathy and more limited in neurodegenerative or treatment-related conditions. The cornea can be considered a surrogate tissue, not the primary site of systemic neuropathy in this regard. Quantification of corneal nerve morphology may allow earlier objective identification of structural small-fiber changes, longitudinal assessment, and multimodal phenotyping of peripheral nerve injury. An abnormal CCM result is not currently sufficient for establishing a diagnosis of systemic neuropathy and does not replace established clinical assessment, quantitative sensory testing, neurophysiological testing, or skin biopsy where indicated. In order to enable broader application in a clinical setting, acquisition and analysis must be standardized, patient-level thresholds validated, multicenter longitudinal validation achieved, and integration with existing diagnostic pathways established. This process may be assisted by automated analysis and AI, but these remain adjunctive pending robust external validation.

Conflict-of-interest statement: All authors declare that they have no conflict of interest to disclose.

AI contribution statement: During the preparation and revision of this manuscript, ChatGPT-5.4 Sol (OpenAI) and Grammarly were used solely to assist with language editing, rephrasing, structural organization, and readability. These tools were not used to generate scientific evidence, perform analyses, determine evidence interpretation, or formulate scientific conclusions. No AI-generated images were used. All factual statements, numerical data, references, interpretations, and conclusions were independently checked and approved by the authors, who take full responsibility for the final manuscript.

Peer review: Externally peer reviewed.

Peer-review model: Single blind

Specialty type: Ophthalmology

Country of origin: Italy

Peer-review report’s classification

Scientific quality: Grade B, Grade C

Novelty: Grade C, Grade C

Creativity or innovation: Grade C, Grade C

Scientific significance: Grade B, Grade C

P-Reviewer: Wang H, Associate Chief Physician, Associate Professor, PhD, China; Zhou X, Assistant Professor, Deputy Director, Vice Director, China S-Editor: Liu JH L-Editor: A P-Editor: Wang WB

1 Azmi S Ferdousi M Petropoulos IN Ponirakis G Alam U Fadavi H Asghar O Marshall A Atkinson AJ Jones W Boulton AJ Tavakoli M Jeziorska M Malik RA Corneal Confocal Microscopy Identifies Small-Fiber Neuropathy in Subjects With Impaired Glucose Tolerance Who Develop Type 2 Diabetes Diabetes Care 2015 38 1502 1508 10.2337/dc14-2733 PMC4512140 25877814

2 Quattrini C Tavakoli M Jeziorska M Kallinikos P Tesfaye S Finnigan J Marshall A Boulton AJ Efron N Malik RA Surrogate markers of small fiber damage in human diabetic neuropathy Diabetes 2007 56 2148 2154 10.2337/db07-0285 17513704

3 Tavakoli M Quattrini C Abbott C Kallinikos P Marshall A Finnigan J Morgan P Efron N Boulton AJ Malik RA Corneal confocal microscopy: a novel noninvasive test to diagnose and stratify the severity of human diabetic neuropathy Diabetes Care 2010 33 1792 1797 10.2337/dc10-0253 PMC2909064 20435796

4 Jiang MS Yuan Y Gu ZX Zhuang SL Corneal confocal microscopy for assessment of diabetic peripheral neuropathy: a meta-analysis Br J Ophthalmol 2016 100 9 14 10.1136/bjophthalmol-2014-306038 25677672

5 Gad H Petropoulos IN Khan A Ponirakis G MacDonald R Alam U Malik RA Corneal confocal microscopy for the diagnosis of diabetic peripheral neuropathy: A systematic review and meta-analysis J Diabetes Investig 2022 13 134 147 10.1111/jdi.13643 PMC8756328 34351711

6 Lim SH Ferdousi M Kalteniece A Kass-Iliyya L Petropoulos IN Malik RA Kobylecki C Silverdale M Corneal confocal microscopy detects small fibre neurodegeneration in Parkinson's disease using automated analysis Sci Rep 2020 10 20147 10.1038/s41598-020-76768-x PMC7677379 33214572

7 Che NN Yang HQ Potential use of corneal confocal microscopy in the diagnosis of Parkinson's disease associated neuropathy Transl Neurodegener 2020 9 28 10.1186/s40035-020-00204-3 PMC7330988 32611440

8 Che NN Chen S Jiang QH Chen SY Zhao ZX Li X Malik RA Ma JJ Yang HQ Corneal confocal microscopy differentiates patients with Parkinson's disease with and without autonomic involvement NPJ Parkinsons Dis 2022 8 114 10.1038/s41531-022-00387-8 PMC9463159 36085290

9 Toprak M Altintas O Bickin H Efendi H Yilmaz Tugan B Yuksel N In vivo confocal microscopy of corneal nerve fiber damage in early course of multiple sclerosis Int Ophthalmol 2023 43 503 509 10.1007/s10792-022-02448-6 35945411

10 Ferrari G Gemignani F Macaluso C Chemotherapy-associated peripheral sensory neuropathy assessed using in vivo corneal confocal microscopy Arch Neurol 2010 67 364 365 10.1001/archneurol.2010.17 20212239

11 Riva N Bonelli F Lasagni Vitar RM Barbariga M Fonteyne P Lopez ID Domi T Scarpa F Ruggeri A Reni M Marcatti M Quattrini A Agosta F Rama P Ferrari G Corneal and Epidermal Nerve Quantification in Chemotherapy Induced Peripheral Neuropathy Front Med (Lausanne) 2022 9 832344 10.3389/fmed.2022.832344 PMC8894874 35252263

12 Mehra S Tavakoli M Kallinikos PA Efron N Boulton AJ Augustine T Malik RA Corneal confocal microscopy detects early nerve regeneration after pancreas transplantation in patients with type 1 diabetes Diabetes Care 2007 30 2608 2612 10.2337/dc07-0870 17623821

13 Gad H Elgassim E Lebbe A MacDonald RS Baraka A Petropoulos IN Ponirakis G Ibrahim NO Malik RA Corneal confocal microscopy detects early nerve regeneration after pharmacological and surgical interventions: Systematic review and meta-analysis J Peripher Nerv Syst 2024 29 173 184 10.1111/jns.12641 38887985

14 Petropoulos IN Alam U Fadavi H Marshall A Asghar O Dabbah MA Chen X Graham J Ponirakis G Boulton AJ Tavakoli M Malik RA Rapid automated diagnosis of diabetic peripheral neuropathy with in vivo corneal confocal microscopy Invest Ophthalmol Vis Sci 2014 55 2071 2078 10.1167/iovs.13-13787 PMC3979234 24569580

15 Chen X Graham J Dabbah MA Petropoulos IN Tavakoli M Malik RA An Automatic Tool for Quantification of Nerve Fibers in Corneal Confocal Microscopy Images IEEE Trans Biomed Eng 2017 64 786 794 10.1109/TBME.2016.2573642 PMC5512547 27295646

16 Williams BM Borroni D Liu R Zhao Y Zhang J Lim J Ma B Romano V Qi H Ferdousi M Petropoulos IN Ponirakis G Kaye S Malik RA Alam U Zheng Y An artificial intelligence-based deep learning algorithm for the diagnosis of diabetic neuropathy using corneal confocal microscopy: a development and validation study Diabetologia 2020 63 419 430 10.1007/s00125-019-05023-4 PMC6946763 31720728

17 Preston FG Meng Y Burgess J Ferdousi M Azmi S Petropoulos IN Kaye S Malik RA Zheng Y Alam U Artificial intelligence utilising corneal confocal microscopy for the diagnosis of peripheral neuropathy in diabetes mellitus and prediabetes Diabetologia 2022 65 457 466 10.1007/s00125-021-05617-x PMC8803718 34806115

18 Petropoulos IN Ponirakis G Khan A Gad H Almuhannadi H Brines M Cerami A Malik RA Corneal confocal microscopy: ready for prime time Clin Exp Optom 2020 103 265 277 10.1111/cxo.12887 30834591

19 Petropoulos IN Ponirakis G Ferdousi M Azmi S Kalteniece A Khan A Gad H Bashir B Marshall A Boulton AJM Soran H Malik RA Corneal Confocal Microscopy: A Biomarker for Diabetic Peripheral Neuropathy Clin Ther 2021 43 1457 1475 10.1016/j.clinthera.2021.04.003 33965237

20 Tavakoli M Ferdousi M Petropoulos IN Morris J Pritchard N Zhivov A Ziegler D Pacaud D Romanchuk K Perkins BA Lovblom LE Bril V Singleton JR Smith G Boulton AJ Efron N Malik RA Normative values for corneal nerve morphology assessed using corneal confocal microscopy: a multinational normative data set Diabetes Care 2015 38 838 843 10.2337/dc14-2311 PMC4407754 25633665

21 Petropoulos IN Manzoor T Morgan P Fadavi H Asghar O Alam U Ponirakis G Dabbah MA Chen X Graham J Tavakoli M Malik RA Repeatability of in vivo corneal confocal microscopy to quantify corneal nerve morphology Cornea 2013 32 e83 e89 10.1097/ICO.0b013e3182749419 23172119

22 Kalteniece A Ferdousi M Adam S Schofield J Azmi S Petropoulos I Soran H Malik RA Corneal confocal microscopy is a rapid reproducible ophthalmic technique for quantifying corneal nerve abnormalities PLoS One 2017 12 e0183040 10.1371/journal.pone.0183040 PMC5560560 28817609

23 Müller LJ Marfurt CF Kruse F Tervo TM Corneal nerves: structure, contents and function Exp Eye Res 2003 76 521 542 10.1016/s0014-4835(03)00050-2 12697417

24 Shaheen BS Bakir M Jain S Corneal nerves in health and disease Surv Ophthalmol 2014 59 263 285 10.1016/j.survophthal.2013.09.002 PMC4004679 24461367

25 Medeiros CS Santhiago MR Corneal nerves anatomy, function, injury and regeneration Exp Eye Res 2020 200 108243 10.1016/j.exer.2020.108243 32926895

26 Yang AY Chow J Liu J Corneal Innervation and Sensation: The Eye and Beyond Yale J Biol Med 2018 91 13 21 PMC5872636 29599653

27 Sacchetti M Lambiase A Neurotrophic factors and corneal nerve regeneration Neural Regen Res 2017 12 1220 1224 10.4103/1673-5374.213534 PMC5607810 28966630

28 Lasagni Vitar RM Rama P Ferrari G The two-faced effects of nerves and neuropeptides in corneal diseases Prog Retin Eye Res 2022 86 100974 10.1016/j.preteyeres.2021.100974 34098111

29 Lukashenko MV Gavrilova NY Bregovskaya AV Soprun LA Churilov LP Petropoulos IN Malik RA Shoenfeld Y Corneal Confocal Microscopy in the Diagnosis of Small Fiber Neuropathy: Faster, Easier, and More Efficient Than Skin Biopsy? Pathophysiology 2021 29 1 8 10.3390/pathophysiology29010001 PMC8954271 35366285

30 Petropoulos IN Madani OA Gad HY Mahfoud ZR Ponirakis G MacDonald R Malik RA Corneal Confocal Microscopy to Diagnose Peripheral Neuropathy: A Systematic Review and Meta-Analysis Eur J Neurol 2025 32 e70396 10.1111/ene.70396 PMC12593550 41201232

31 Mizukami H Osonoi S Pathogenesis and Molecular Treatment Strategies of Diabetic Neuropathy Collateral Glucose-Utilizing Pathways in Diabetic Polyneuropathy Int J Mol Sci 2020 22 94 10.3390/ijms22010094 PMC7796340 33374137

32 Pang L Lian X Liu H Zhang Y Li Q Cai Y Ma H Yu X Understanding Diabetic Neuropathy: Focus on Oxidative Stress Oxid Med Cell Longev 2020 2020 9524635 10.1155/2020/9524635 PMC7422494 32832011

33 Zhu J Hu Z Luo Y Liu Y Luo W Du X Luo Z Hu J Peng S Diabetic peripheral neuropathy: pathogenetic mechanisms and treatment Front Endocrinol (Lausanne) 2023 14 1265372 10.3389/fendo.2023.1265372 PMC10803883 38264279

34 Eid SA Rumora AE Beirowski B Bennett DL Hur J Savelieff MG Feldman EL New perspectives in diabetic neuropathy Neuron 2023 111 2623 2641 10.1016/j.neuron.2023.05.003 PMC10525009 37263266

35 Baum P Toyka KV Blüher M Kosacka J Nowicki M Inflammatory Mechanisms in the Pathophysiology of Diabetic Peripheral Neuropathy (DN)-New Aspects Int J Mol Sci 2021 22 10835 10.3390/ijms221910835 PMC8509236 34639176

36 Klemmensen MM Borrowman SH Pearce C Pyles B Chandra B Mitochondrial dysfunction in neurodegenerative disorders Neurotherapeutics 2024 21 e00292 10.1016/j.neurot.2023.10.002 PMC10903104 38241161

37 Berth SH Lloyd TE Disruption of axonal transport in neurodegeneration J Clin Invest 2023 133 e168554 10.1172/JCI168554 PMC10232001 37259916

38 Kalteniece A Ferdousi M Azmi S Mubita WM Marshall A Lauria G Faber CG Soran H Malik RA Corneal confocal microscopy detects small nerve fibre damage in patients with painful diabetic neuropathy Sci Rep 2020 10 3371 10.1038/s41598-020-60422-7 PMC7042367 32099076

39 Petroll WM Robertson DM In Vivo Confocal Microscopy of the Cornea: New Developments in Image Acquisition, Reconstruction, and Analysis Using the HRT-Rostock Corneal Module Ocul Surf 2015 13 187 203 10.1016/j.jtos.2015.05.002 PMC4499020 25998608

40 Cañadas P Alberquilla García-Velasco M Hernández Verdejo JL Teus MA Update on Corneal Confocal Microscopy Imaging Diagnostics (Basel) 2022 13 46 10.3390/diagnostics13010046 PMC9818591 36611338

41 Alam U Anson M Meng Y Preston F Kirthi V Jackson TL Nderitu P Cuthbertson DJ Malik RA Zheng Y Petropoulos IN Artificial Intelligence and Corneal Confocal Microscopy: The Start of a Beautiful Relationship J Clin Med 2022 11 6199 10.3390/jcm11206199 PMC9604848 36294519

42 Kheirkhah A Muller R Mikolajczak J Ren A Kadas EM Zimmermann H Pruess H Paul F Brandt AU Hamrah P Comparison of Standard Versus Wide-Field Composite Images of the Corneal Subbasal Layer by In Vivo Confocal Microscopy Invest Ophthalmol Vis Sci 2015 56 5801 5807 10.1167/iovs.15-17434 PMC4559214 26325419

43 Schenk MS Wartak A Buehler V Zhao J Tearney GJ Birngruber R Kassumeh S Advances in Imaging of Subbasal Corneal Nerves With Micro-Optical Coherence Tomography Transl Vis Sci Technol 2021 10 22 10.1167/tvst.10.13.22 PMC8606792 34779835

44 Labetoulle M Baudouin C Calonge M Merayo-Lloves J Boboridis KG Akova YA Aragona P Geerling G Messmer EM Benítez-Del-Castillo J Role of corneal nerves in ocular surface homeostasis and disease Acta Ophthalmol 2019 97 137 145 10.1111/aos.13844 30225941

45 Chiang JCB Roy M Kim J Markoulli M Krishnan AV In-vivo corneal confocal microscopy: Imaging analysis, biological insights and future directions Commun Biol 2023 6 652 10.1038/s42003-023-05005-8 PMC10279669 37336941

46 Kryszan K Wylęgała A Kijonka M Potrawa P Walasz M Wylęgała E Orzechowska-Wylęgała B Artificial-Intelligence-Enhanced Analysis of In Vivo Confocal Microscopy in Corneal Diseases: A Review Diagnostics (Basel) 2024 14 694 10.3390/diagnostics14070694 PMC11011861 38611606

47 Gylfadottir SS Itani M Kristensen AG Nyengaard JR Sindrup SH Jensen TS Finnerup NB Karlsson P Assessing Corneal Confocal Microscopy and Other Small Fiber Measures in Diabetic Polyneuropathy Neurology 2023 100 e1680 e1690 10.1212/WNL.0000000000206902 PMC10115507 36750383

48 Perkins BA Lovblom LE Bril V Scarr D Ostrovski I Orszag A Edwards K Pritchard N Russell A Dehghani C Pacaud D Romanchuk K Mah JK Jeziorska M Marshall A Shtein RM Pop-Busui R Lentz SI Boulton AJM Tavakoli M Efron N Malik RA Corneal confocal microscopy for identification of diabetic sensorimotor polyneuropathy: a pooled multinational consortium study Diabetologia 2018 61 1856 1861 10.1007/s00125-018-4653-8 PMC6061173 29869146

49 Wang M Zhang C Zuo A Li L Chen L Hou X Diagnostic utility of corneal confocal microscopy in type 2 diabetic peripheral neuropathy J Diabetes Investig 2021 12 574 582 10.1111/jdi.13381 PMC8015829 32745370

50 Badian RA Ekman L Pripp AH Utheim TP Englund E Dahlin LB Rolandsson O Lagali N Comparison of Novel Wide-Field In Vivo Corneal Confocal Microscopy With Skin Biopsy for Assessing Peripheral Neuropathy in Type 2 Diabetes Diabetes 2023 72 908 917 10.2337/db22-0863 PMC10281223 37058418

51 Bjørnkaer A Gaist LM Holbech JV Gaist D Wirenfeldt M Sindrup SH Krøigård T Corneal confocal microscopy in small and mixed fiber neuropathy-Comparison with skin biopsy and cold detection in a large prospective cohort J Peripher Nerv Syst 2023 28 664 676 10.1111/jns.12595 37651181

52 Sadhu S Jalbert I Palacios-Derflingher L Alghamdi A Golebiowski B Stapleton F Ocular surface epithelial immune cells and corneal nerves in dry eye disease - A systematic review and meta-analysis of in vivo confocal microscopy data Ocul Surf 2025 38 80 96 10.1016/j.jtos.2025.06.005 40545036

53 Bandeira F Yusoff NZ Yam GH Mehta JS Corneal re-innervation following refractive surgery treatments Neural Regen Res 2019 14 557 565 10.4103/1673-5374.247421 PMC6352585 30632489

54 Giannaccare G Bernabei F Pellegrini M Guaraldi F Turchi F Torrazza C Senni C Scotto R Sindaco D Di Cello L Versura P Scorcia V Traverso CE Vagge A Bilateral morphometric analysis of corneal sub-basal nerve plexus in patients undergoing unilateral cataract surgery: a preliminary in vivo confocal microscopy study Br J Ophthalmol 2021 105 174 179 10.1136/bjophthalmol-2019-315449 32245849

55 Chiang JCB Tajbakhsh Z Wolffsohn JS The clinical impact of contact lens wear on neural structure and function of the cornea Clin Exp Optom 2025 108 408 418 10.1080/08164622.2024.2401511 39250904

56 Collins GS Moons KGM Dhiman P Riley RD Beam AL Van Calster B Ghassemi M Liu X Reitsma JB van Smeden M Boulesteix AL Camaradou JC Celi LA Denaxas S Denniston AK Glocker B Golub RM Harvey H Heinze G Hoffman MM Kengne AP Lam E Lee N Loder EW Maier-Hein L Mateen BA McCradden MD Oakden-Rayner L Ordish J Parnell R Rose S Singh K Wynants L Logullo P TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods BMJ 2024 385 e078378 10.1136/bmj-2023-078378 PMC11019967 38626948

57 Moons KGM Damen JAA Kaul T Hooft L Andaur Navarro C Dhiman P Beam AL Van Calster B Celi LA Denaxas S Denniston AK Ghassemi M Heinze G Kengne AP Maier-Hein L Liu X Logullo P McCradden MD Liu N Oakden-Rayner L Singh K Ting DS Wynants L Yang B Reitsma JB Riley RD Collins GS van Smeden M PROBAST+AI: an updated quality, risk of bias, and applicability assessment tool for prediction models using regression or artificial intelligence methods BMJ 2025 388 e082505 10.1136/bmj-2024-082505 PMC11931409 40127903