Conjunctival Bulbar Redness Extraction Pipeline for High-Resolution Ocular Surface Photography.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Conjunctival bulbar redness is a key indicator of ocular surface inflammation.
• Objective and reproducible grading of ocular redness is crucial for clinical decision-making.
• Current methods for assessing redness can be subjective and time-consuming.

Purpose of the Study:

• To develop and validate an image analysis pipeline for extracting conjunctival bulbar redness from standardized ocular surface photographs.
• To implement a machine learning model for accurate ocular surface segmentation.
• To establish an objective, image-based redness score for clinical use.

Main Methods:

• Collected data from healthy individuals and ophthalmic clinic patients using a novel imaging system.
• Trained a machine learning model for ocular surface segmentation and redness biomarker extraction.
• Correlated image-based redness scores with established clinical gradings (Efron) for validation.

Main Results:

• Achieved high segmentation performance with mean intersection over union of 0.9639 (iris) and 0.9731 (ocular surface).
• Established a digital grading scale and demonstrated good feasibility and reproducibility of redness scores.
• Found a moderate positive correlation (Spearman's rho = 0.599) between image-based scores and clinical grading.

Conclusions:

• The developed pipeline enables objective classification and evaluation of ocular redness using standardized external eye photography.
• This AI-driven approach offers a potential objective tool for grading ocular redness in a high-throughput manner.
• The system can empower eye care professionals with data to facilitate clinical decision-making.