An Artificial Intelligence Enabled System for Retinal Nerve Fiber Layer Thickness Damage Severity Staging.

Area of Science:

• Ophthalmology
• Medical Imaging
• Machine Learning

Background:

• Glaucoma is a leading cause of irreversible blindness.
• Accurate glaucoma damage severity classification is crucial for patient management.
• Current methods may lack objectivity and consistency.

Purpose of the Study:

• To develop an objective glaucoma damage severity classification system.
• Utilize optical coherence tomography (OCT)-derived retinal nerve fiber layer (RNFL) thickness measurements.
• Establish a classification system based on unsupervised learning.

Main Methods:

• Developed an unsupervised k-means model using multicenter OCT data.
• Clustered eyes based on circumpapillary RNFL thickness profiles.
• Determined optimal RNFL thickness thresholds using Bayes' minimum error principle for severity classification.

Main Results:

• Identified four distinct clusters representing different RNFL thickness profiles.
• Established optimal global RNFL thickness thresholds: >95 μm (normal), 86-95 μm (early), 70-85 μm (moderate), <70 μm (advanced).
• Achieved high accuracy in classifying glaucoma severity, with 98% of advanced cases correctly identified.

Conclusions:

• An objective, unbiased, and consistent glaucoma damage classification system was developed.
• The system utilizes OCT-derived RNFL thickness measurements and unsupervised machine learning.
• This classification method can enhance glaucoma research and clinical practice.