An artificial intelligence-based deep learning algorithm for the diagnosis of diabetic neuropathy using corneal...

Area of Science:

• Ophthalmology
• Neuroscience
• Medical Imaging

Background:

• Corneal confocal microscopy aids in identifying neurodegenerative diseases.
• Quantifying corneal sub-basal nerve plexus morphology is crucial but challenging.
• Current methods involve time-consuming manual analysis or less sensitive automated approaches.

Purpose of the Study:

• To develop and validate an AI-based deep learning algorithm for quantifying corneal nerve fiber properties.
• To assess the algorithm's utility in diagnosing diabetic neuropathy.
• To compare the algorithm's performance against a validated automated analysis program (ACCMetrics).

Main Methods:

• A convolutional neural network with data augmentation was employed for automated quantification.
• The algorithm was trained on 1698 images and validated on 2137 images.
• Key nerve fiber properties quantified include total length, branch points, and fractal numbers.

Main Results:

• The deep learning algorithm demonstrated superior intraclass correlation coefficients compared to ACCMetrics for most parameters.
• It achieved an AUC of 0.83, with 0.87 specificity and 0.68 sensitivity for classifying diabetic neuropathy.
• The algorithm provides rapid and excellent localization performance for corneal nerve biomarkers.

Conclusions:

• The developed deep learning algorithm offers a significant advancement in quantifying corneal nerve morphology.
• It shows high accuracy and efficiency, outperforming existing automated methods.
• The algorithm holds potential for integration into clinical screening programs for diabetic neuropathy.