Machine Learning Models for Diagnosing Glaucoma from Retinal Nerve Fiber Layer Thickness Maps.

Area of Science:

• Ophthalmology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Glaucoma diagnosis traditionally relies on limited structural measurements.
• Retinal nerve fiber layer (RNFL) thickness maps contain rich spatial information crucial for detecting glaucomatous damage.
• Machine learning (ML) offers potential for enhanced diagnostic accuracy by analyzing complex patterns in medical data.

Purpose of the Study:

• To evaluate the diagnostic accuracy of multiple ML models in detecting glaucoma.
• To compare the performance of ML models utilizing full RNFL thickness maps against conventional metrics.

Main Methods:

• A case-control study involving 93 eyes with glaucoma and 128 healthy control eyes from the Los Angeles Latino Eye Study (LALES).
• Four ML algorithms were applied to 6x6-mm RNFL thickness maps: Support Vector Machine (SVM), K-Nearest Neighbor (KNN), ResNet-18, and a custom GlaucomaNet.
• Model performance was assessed using Area Under the Curve (AUC) via 5-fold cross-validation, compared to average circumpapillary RNFL thickness.

Main Results:

• All four ML models demonstrated high diagnostic accuracies, with AUC values ranging from 0.91 to 0.92.
• These AUC values were significantly higher than the AUC of 0.76 obtained using conventional average circumpapillary RNFL thickness.
• Both conventional ML and deep learning (convolutional neural net) models outperformed the standard measurement.

Conclusions:

• ML models leveraging full RNFL thickness maps provide superior diagnostic performance for glaucoma detection.
• The spatial structure within RNFL thickness map data is critical for accurate glaucoma diagnosis.
• Further research should focus on optimizing the utilization of comprehensive RNFL map data for improved clinical application.