A Novel Interactive, Artificial Intelligence (Ai) Platform For Real-Time Analysis Of Normal Characteristics And Morphology Of Corneal Subbasal Nerve Plexus.

Published 2026 - 44th Congress of the ESCRS

Reference: POCOR532 | Type: Free paper | DOI: 10.82333/p7q0-gw17

Authors: Prof. Dr Manuel Ramirez* ¹ , Dr Gustavo Rubio ¹ , Dr Emmanuel Cabrera ¹ , Dr Everardo Hernandez-Quintela ² , Itzel Aguilar ¹ , Dr Alejandro Babayan ³

¹Cornea Department,APEC,Mexico City,Mexico, ²Ophthalmology Department, Wilmer Eye Institute, Johns Hopkins University School of Medicine,Baltimore,United States, ³Cornea Department,Hospital de la Luz,Mexico City,Mexico

Purpose

To develop and validate an accessible, web-based platform that combines automated AI analysis in real-time and expert-user correction to accurately evaluate subbasal nerve plexus (SBNP) characteristics and morphology from in vivo corneal confocal microscopy (IVCM) images.

Setting

The study was developed in a hight referral ophthalmology hospital in Mexico City, Mexico

Methods

400 SBNP images from 200 healthy patients’ corneal in vivo confocal microscopy scans were analyzed (94 men, 106 women; mean age was 31.5 ± 6.5, range 21–50 years). The best in focus SBNP image of each corneal confocal microscopy scan was selected. A browser-based application was developed featuring a three-stage workflow: 1) Initial manual tracing, 2) Semi-automated training with "human-in-the-loop" validation, and 3) Analysis and Reinforcement.

Results

We successfully generated an artificial intelligence model. It provided real-time quantification of NFD and nerve plexus orientation with visual feedback (heatmaps). User controls, such as manual orientation overrides and sensitivity adjustment by using the software settings, allowed for expert correction of inaccuracies. In testing with the 400-image dataset, the model confidence score demonstrated a clear positive trend, indicating improved consistency and reduced data variance.

Conclusions

We successfully developed an interactive platform bridging the gap between manual analysis and inflexible previous semi-automated systems. Artificial Intelligence proved to be an effective and efficient resource for obtaining SBNP characteristics, with the potential for future improvement. The "Add to Model" feature and manual overrides ensure expert-level accuracy, making this a powerful tool for high-throughput analysis in corneal neuropathy studies.