Prediction of Sex and Age from Macular Optical Coherence Tomography Images and Feature Analysis Using Deep Learning

The prevalence of certain macular diseases differs between male and female. However, the actual difference in macular structure between male and female was barely understood. Previous studies reported the mean retinal thickness of macula was thinner for female, but here it was observed that the difference is not statistically large enough for sex distinction. Similarly, the age-related non-pathological change of macular structure was also hardly known. It has been found that the thickness of choroid decreases with age. In this study, deep learning was applied to distinguish sex and age from macular optical coherence tomography (OCT) images of 3134 persons and achieved a sex prediction accuracy of 85.6 ± 2.1% and an age prediction error of 5.78 ± 0.29 years. A thorough analysis of the prediction accuracy and the Grad-CAM showed that 1) the foveal contour leads to a better sex distinction than the macular thickness, 2) B-scan macular OCT images contain more sex-related information than en face fundus images, and 3) the age-related characteristics of the macula are on the whole layers of the retina, not just the choroid. These novel findings reported in this study are useful to ophthalmologists for further investigation in the pathogenesis of sex and age-related macular structural diseases.