Prediction of Axial Length From Macular Optical Coherence Tomography Using Deep Learning Model

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate measurement of axial length (AL) is crucial for ophthalmic diagnostics and refractive error prediction.
• Optical coherence tomography (OCT) provides high-resolution cross-sectional images of the retina and choroid.

Purpose of the Study:

• To develop and evaluate a deep learning model for predicting axial length (AL) from optical coherence tomography (OCT) images.
• To assess the performance of different OCT image input configurations (horizontal, vertical, dual-input) for AL prediction.

Main Methods:

• Retrospective analysis of 9064 eyes with AL measurements and OCT images.
• Utilized ResNet-152 architecture with 5-fold cross-validation.
• Trained models using horizontal OCT, vertical OCT, and dual-input (horizontal + vertical) images.

Main Results:

• The dual-input deep learning model achieved a mean absolute error (MAE) of 0.592 mm and R-squared (R2) of 0.847 in the internal test set.
• External test set validation showed MAE of 0.556 mm and R2 of 0.663.
• Dual-input models achieved >83% accuracy within ±1.0 mm error margin in both test sets.

Conclusions:

• A deep learning-based model can accurately predict axial length from OCT images.
• The dual-input model, utilizing both horizontal and vertical OCT images, demonstrated the best predictive performance.
• Macular OCT images hold significant potential for AL prediction using AI.