TL;DR
This study leverages corneal confocal microscopy (CCM) and two advanced AI-driven image analysis algorithms to demonstrate significant axonal loss in the cornea of multiple sclerosis (MS) patients across various disease stages, including early clinically isolated syndrome (CIS). This non-invasive technique shows promise as a sensitive biomarker for neurodegeneration in MS, correlating with disease severity.
What Problem Does This Paper Address?
Multiple sclerosis (MS) is a neurodegenerative disease where axonal loss is a primary driver of progression and disability. Accurately quantifying this loss is crucial for predicting patient outcomes and evaluating the efficacy of new neuroprotective therapies. While traditional methods like MRI and OCT have been explored, they have limitations. This paper investigates whether corneal confocal microscopy (CCM), a non-invasive technique already established for detecting peripheral neuropathies, can serve as a sensitive and reliable tool to detect and quantify axonal loss in the central nervous system disease, MS, across its different clinical courses. A key question is whether this axonal loss is detectable even in the earliest stages of MS (clinically isolated syndrome).
Methods
The study employed corneal confocal microscopy (CCM) to image the sub-basal nerve plexus of 72 MS patients (across clinically isolated syndrome, relapsing-remitting MS, and secondary-progressive MS) and 21 healthy controls. The novelty lies in the use of two independent, automated image segmentation algorithms to quantify corneal nerve parameters:
- ACCMetrics: A machine-learning-based algorithm previously validated for diabetic neuropathy. It quantifies corneal nerve fiber density (CNFD), corneal nerve fiber length (CNFL), and corneal nerve fractal dimension (CNFrD).
- Voxeleron deepNerve: A deep learning-based algorithm that uses a U-Net architecture. It quantifies CNFL and CNFrD.
The use of two distinct algorithms, one traditional machine learning and one deep learning, was crucial for validating the findings and demonstrating the robustness of the observed axonal loss. Reproducibility between the algorithms was also assessed.
Datasets
The study utilized a dataset of 144 eyes from 72 human subjects:
- Clinically Isolated Syndrome (CIS): n=9
- Relapsing-Remitting MS (RRMS): n=20
- Secondary-Progressive MS (SPMS): n=22
- Age-matched Healthy Controls (HC): n=21
No new benchmark datasets were introduced; the study focused on applying existing and novel analysis techniques to a well-defined patient cohort.
Key Insights and Ideas
- Corneal Axonal Loss is an Early MS Biomarker: The most significant finding is the demonstration of substantial corneal axonal loss not only in established MS but also in patients with clinically isolated syndrome (CIS), suggesting it's an early manifestation of neurodegeneration in MS.
- CCM as a Sensitive Tool for MS Neurodegeneration: The study validates CCM as a sensitive, non-invasive method to detect and quantify axonal loss, which is a key determinant of MS progression.
- AI-Driven Analysis Enhances Reliability: The use of two independent, automated segmentation algorithms (ACCMetrics and Voxeleron deepNerve) provides strong evidence for the reliability and reproducibility of the findings, overcoming the subjectivity and time constraints of manual analysis.
- Corneal Nerve Loss Correlates with Disability: The study found significant correlations between corneal nerve parameters (CNFD, CNFL, CNFrD) and the Expanded Disability Status Scale (EDSS), reinforcing the link between corneal nerve integrity and neurological impairment in MS.
- Potential for Therapeutic Monitoring: The ability to quantify axonal loss non-invasively opens doors for using CCM in clinical trials to monitor the effectiveness of neuroprotective therapies.
Key Results
- Significant Reduction in Corneal Nerve Parameters: Patients with MS, across all disease courses (CIS, RRMS, SPMS), showed significantly lower CNFD, CNFL, and CNFrD compared to healthy controls.
- CNFD (ACCMetrics): 19.85 ± 6.75 fibers/mm² in MS vs. 34.76 ± 5.57 fibers/mm² in controls (P < .0001).
- CNFL (ACCMetrics): 12.40 ± 3.30 mm/mm² in MS vs. 19.75 ± 2.39 mm/mm² in controls (P < .0001).
- CNFL (Voxeleron deepNerve): 14.40 ± 4.17 mm/mm² in MS vs. 21.98 ± 2.76 mm/mm² in controls (P < .0001).
- CNFrD (ACCMetrics): 1.45 ± 0.04 in MS vs. 1.52 ± 0.02 in controls (P < .0001).
- CNFrD (Voxeleron deepNerve): 1.19 ± 0.07 in MS vs. 1.29 ± 0.03 in controls (P < .0001).
- Comparable Loss Across MS Courses: Corneal nerve parameters were comparably reduced across CIS, RRMS, and SPMS, suggesting that axonal loss occurs early and persists throughout the disease.
- Excellent Reproducibility: The two independent algorithms demonstrated excellent reproducibility, with strong linear relationships and high agreement (Bland-Altman analysis).
- EDSS Correlation: EDSS scores correlated significantly with CNFD (r=-0.32), CNFL (r=-0.32), and CNFrD (r=-0.29).
Limitations and Future Work
- Image Sampling Bias: The study acknowledges that the image sampling approach might have introduced selection bias, potentially favoring lower mean values, especially in patients with more advanced disability.
- Generalizability of Deep Learning: The results from the Voxeleron deepNerve algorithm may not be directly generalizable to all deep learning algorithms due to variations in network architecture.
- Trigeminal Neuralgia Contribution: While no participants had a clinical diagnosis of trigeminal neuralgia, the possibility that trigeminal-related pathology might contribute to observed differences is noted. However, prior studies suggest no significant difference in corneal nerve density in patients with or without trigeminal neuralgia-related symptoms.
- Future Directions: The authors suggest further prospective CCM studies in larger cohorts across different MS courses. They also recommend evaluating additional morphological features (e.g., inferior whorl, Langerhans cells) and exploring these findings longitudinally in relation to trigeminal neuralgia and therapeutic interventions. The potential of CCM as a rapid, non-invasive surrogate marker for neurodegeneration in MS is highlighted.
Summary updated