Estimation of foveal avascular zone area from a B-scan OCT image using machine learning algorithms

Area of Science:

• Ophthalmology
• Medical Imaging
• Machine Learning

Background:

• The Foveal Avascular Zone (FAZ) is a critical indicator of retinal health.
• Estimating FAZ area is essential for diagnosing and monitoring retinal vascular diseases.
• Optical Coherence Tomography (OCT) B-scan images offer a potential source for FAZ analysis.

Purpose of the Study:

• To develop and evaluate machine learning models for estimating FAZ area using OCT B-scan images.
• To assess the accuracy and reliability of these models in predicting FAZ dimensions.

Main Methods:

• Developed three machine learning models to predict FAZ length, estimate FAZ area from length using multiple measurements, and convert pixel measurements to mm².
• Evaluated model performance using Mean Absolute Error (MAE), Mean Squared Error (MSE), and Coefficient of Determination (R2).
• Applied models sequentially to a new dataset to estimate FAZ area from OCT B-scan images.

Main Results:

• Model 1 (FAZ length prediction) achieved R2 of 0.87.
• Model 2 (FAZ area estimation) showed improved accuracy with more measurements, reaching R2 of 0.95 with 5 lines.
• Model 3 (pixel to mm² conversion) demonstrated high accuracy (R2=1.0), and overall FAZ area estimation accuracy increased with more B-scan images.

Conclusions:

• Successfully developed machine learning models to predict FAZ area from OCT B-scan images.
• These models hold potential for OCT angiography (OCTA) data prediction, especially in pediatric and geriatric populations.
• Future research can leverage these findings to study FAZ and macular development and retinal health.