Programa del congreso

The brain-age delta, defined as the difference between an individual’s chronological age and the brain age estimated using structural neuroimaging, has been proposed as a marker of biological brain aging. In the context of Alzheimer’s disease (AD), this paradigm has been recently validated against biomarkers of AD pathology and neurodegeneration in non-demented indi-viduals. Moreover, it has also been shown to mediate the relationship be-tween modifiable risk factors and longitudinal cognitive function in mid-dle-aged cognitively unimpaired individuals. Nevertheless, the brain-age delta compresses a rich neuroanatomical heterogeneity – biological, envi-ronmental - in a single objective measurement. Generative deep learning models, such as variational autoencoders (VAE), can learn abstract repre-sentations from neuroimaging data. Using the EPAD and ALFA+ cohorts, we trained several VAE architectures, and utilized their latent spaces to characterize brain aging patterns associated with modifiable risk factors and biomarkers of AD and neurodegeneration. A cognitively “resilient” and a “vulnerable” subtype were found. Moreover, supervised regression models to estimate the brain-age delta were included in the VAE architectures, achieving similar accuracy to previously validated machine learning meth-ods. Therefore, we obtained a complete picture of brain aging, serving as a first step towards tailored interventions for the prevention of cognitive de-cline and dementia.