Investigate Sex Dimorphism of Cerebral Myelination Across Lifespan by Leveraging Conditional Variational Autoencoder

Jinghang Li, Linghang Wang, Chang-le Chen, Tamer Ibrahim, Howard Aizenstein, Minjie Wu

Keywords: Conditional VAE, Brain age prediction, Sex differences, Myelin map

TL;DR: Conditional variational autoencoder trained in a semi-supervised fashion is able to learn the white matter aging differences across sex.

Abstract: In this work we investigated the potential sex differences in white matter aging using conditional variational autoencoder (cVAE) on myelin content MR images. The cVAE model was trained along with a supervised brain age prediction model, which learns the representation of myelination aging process within a single end-to-end model architecture. The training was conducted on a normal aging dataset (CamCAN) that included 708 individual MR images. Our brief exploration revealed that women might have slightly less white matter myelination than men do at an older age. Additionally, our brain age prediction model suggested different aging regressions for men and women.