Artificial Intelligence and Corneal Confocal Microscopy: The Start of a Beautiful Relationship

Uazman Alam, Matthew Anson, Yanda Meng, Frank Preston, Varo Kirthi, Timothy L. Jackson, Paul Nderitu, Daniel J. Cuthbertson, Rayaz A. Malik, Yalin Zheng, Ioannis N. PetropoulosJournal of Clinical Medicine 2022 · DOI

TL;DR

This review explores the burgeoning synergy between Artificial Intelligence (AI), particularly deep learning algorithms (DLAs), and Corneal Confocal Microscopy (CCM). By leveraging AI to analyze complex corneal nerve images, researchers are developing more accurate and efficient diagnostic tools for peripheral neuropathies and neurodegenerative diseases, potentially revolutionizing early detection and patient management.

What Problem Does This Paper Address?

Diagnosing peripheral neuropathies, especially early-stage diabetic peripheral neuropathy (DPN), is challenging. Traditional methods like questionnaires, sensory testing, and nerve conduction studies have limitations in accuracy, reproducibility, or their ability to detect subtle nerve damage. Corneal confocal microscopy (CCM) offers a non-invasive way to visualize corneal nerve fibers, which are known to be affected in various neuropathies. However, manual analysis of CCM images is time-consuming and requires significant expertise, while existing automated methods may lack the necessary precision. This paper addresses the need for more robust, objective, and scalable methods to analyze CCM data for improved diagnostic capabilities, particularly in differentiating between various neuropathies and neurodegenerative conditions.

Methods

The paper highlights the application of Artificial Intelligence (AI), specifically Deep Learning Algorithms (DLAs), to analyze Corneal Confocal Microscopy (CCM) images. DLAs, particularly Convolutional Neural Networks (CNNs), are adept at learning complex patterns directly from image data. The authors discuss how these algorithms can automate the segmentation and quantification of corneal nerve morphology, including parameters like corneal nerve fiber length (CNFL), density (CNFD), and branch density (CNBD). The review emphasizes the shift towards "end-to-end" classification models, where DLAs directly predict disease presence or type from raw images, bypassing traditional feature extraction steps. Techniques like generative adversarial networks (GANs) for image enhancement and attention mechanisms for identifying relevant image regions are also mentioned as advancements in AI-driven CCM analysis.

Datasets

The paper references numerous studies that have utilized existing datasets of CCM images from various patient populations, including healthy volunteers, individuals with type 1 and type 2 diabetes (with and without DPN), and patients with other peripheral neuropathies and central neurodegenerative diseases. While specific details on the exact size and composition of all these datasets are not consolidated into a single benchmark, the review points to the need for large, representative "real-world" datasets for robust AI model development and validation, especially for gaining regulatory approval.

Key Insights and Ideas

  • Corneal nerves as a "window" to neurological health: The cornea's dense innervation makes it a unique and accessible site for assessing peripheral and even central nervous system health non-invasively.
  • AI's potential to overcome manual analysis limitations: DLAs can process vast amounts of CCM data much faster and more consistently than human experts, reducing subjectivity and increasing throughput.
  • End-to-end learning for superior performance: Moving beyond manual feature extraction, DLAs can learn the most discriminative features directly from images, leading to improved diagnostic accuracy.
  • Beyond DPN: Expanding AI's scope: The review strongly advocates for applying AI-driven CCM analysis to differentiate between various peripheral neuropathies and central neurodegenerative diseases, which currently pose significant diagnostic challenges.
  • The "black box" problem and explainability: While AI models show promise, their "black-box" nature is a concern. The paper touches upon the importance of explainability techniques (e.g., heatmaps) to understand why an AI makes a particular diagnosis, building trust and aiding clinical adoption.

Key Results

  • DLAs have demonstrated performance comparable to manual analysis and superior to traditional automated methods for corneal nerve morphology quantification.
  • Recent end-to-end classification models have achieved high sensitivity and specificity in differentiating healthy individuals from those with peripheral neuropathy.
  • Studies using AI have shown excellent correlation with manual quantification of corneal nerve parameters, with some achieving AUCs of 1.0 and sensitivities/specificities of 0.95-1.0 for DPN detection.
  • AI models have shown promise in classifying participants into control, DPN-ve, and DPN+ve categories with high accuracy (e.g., 96% classification accuracy reported).

Limitations and Future Work

  • Data Size and Validation: Current AI models for CCM analysis have primarily been trained and validated on relatively small datasets. Large-scale, external, and prospective clinical validation is crucial for widespread adoption and regulatory approval.
  • Explainability: The "black-box" nature of DLAs remains a challenge. Further development of explainability tools is needed to build clinical trust and facilitate understanding of diagnostic decisions.
  • Regulatory Hurdles: CCM, when used with AI for diagnosis, will likely require regulatory approval as a medical device, necessitating rigorous validation and standardization.
  • Future Directions: The authors envision AI-driven CCM analysis becoming a powerful tool for differentiating between various peripheral neuropathies and central neurodegenerative diseases. They suggest exploring novel corneal nerve parameters and assessing nerve complexity to identify unique disease-specific topographical signatures. The development of a "hub and spoke" model, similar to diabetic retinopathy screening, where AI centrally interprets images, is also proposed to standardize analysis and improve accuracy.

Summary updated

Abstract

Corneal confocal microscopy (CCM) is a rapid non-invasive in vivo ophthalmic imaging technique that images the cornea. Historically, it was utilised in the diagnosis and clinical management of corneal epithelial and stromal disorders. However, over the past 20 years, CCM has been increasingly used to image sub-basal small nerve fibres in a variety of peripheral neuropathies and central neurodegenerative diseases. CCM has been used to identify subclinical nerve damage and to predict the development of diabetic peripheral neuropathy (DPN). The complex structure of the corneal sub-basal nerve plexus can be readily analysed through nerve segmentation with manual or automated quantification of parameters such as corneal nerve fibre length (CNFL), nerve fibre density (CNFD), and nerve branch density (CNBD). Large quantities of 2D corneal nerve images lend themselves to the application of artificial intelligence (AI)-based deep learning algorithms (DLA). Indeed, DLA have demonstrated performance comparable to manual but superior to automated quantification of corneal nerve morphology. Recently, our end-to-end classification with a 3 class AI model demonstrated high sensitivity and specificity in differentiating healthy volunteers from people with and without peripheral neuropathy. We believe there is significant scope and need to apply AI to help differentiate between peripheral neuropathies and also central neurodegenerative disorders. AI has significant potential to enhance the diagnostic and prognostic utility of CCM in the management of both peripheral and central neurodegenerative diseases.
BibTeX
@article{alam2022artificial,
  author = {Uazman Alam and Matthew Anson and Yanda Meng and Frank Preston and Varo Kirthi and Timothy L. Jackson and Paul Nderitu and Daniel J. Cuthbertson and Rayaz A. Malik and Yalin Zheng and Ioannis N. Petropoulos},
  title = {{Artificial Intelligence and Corneal Confocal Microscopy: The Start of a Beautiful Relationship}},
  journal = {Journal of Clinical Medicine},
  year = {2022},
  doi = {10.3390/jcm11206199},
}
Figure