Deep Learning-based age prediction models from retinal Optical Coherence Tomography images

Garazi Zuazo Atutxa, Unai Ayala, Iñigo Gabilondo, Ane Murueta-Goyena, Sara Teijeira, Marta Galdos, Maitane Barrenechea

Keywords: Optical Coherence Tomography, Deep Learning, Age Prediction

Abstract: Optical Coherence Tomography (OCT) has emerged as a valuable tool for assessing microstructural characteristics in the retina. This study explores the use of Deep Learning models for age prediction based on OCT images. We employed three pre-trained ResNet architectures (ResNet-18, ResNet-50, ResNet-152) to predict retinal age from macula raster and peripapillary scans of 517 control subjects. The best performance was achieved by ResNet-18 applied to macula B-scan 12, yielding a Mean Absolute Error (MAE) of 4.423 years. These findings suggest that central macula raster scans, particularly B-scan 12, provide informative features about age-related changes.

Submission Number: 27