Clinical Neuroscience
Aging and Alzheimer’s Disease
Our laboratory was among the first labs to develop machine learning-based imaging biomarkers [Davatzikos et.al., Archives of General Psychiatry, 2004] with emphasis on neuropsychiatry, aging and Alzheimer’s Disease, and to show that these individualized biomarkers have high prognostic value in terms of progressing from normal cognition to mild cognitive impairment, and from MCI to AD. More recently, application of Smile-GAN [Yang et.al., Nature Communications, 2021] (heterogeneity analysis method) revealed four neurodegenerative patterns and two progression pathways of MCI/AD, which imply distinct cognitive performances and predict longitudinal clinical progressions.
Smile-GAN (SeMI-supervised cLustEring-Generative Adversarial Network) is a deep learning method designed for dissecting phenotypic heterogeneity of brain diseases and disorders. It utilizes generative adversarial deep learning methods (GAN) to estimate heterogeneous transformations from a (e.g. healthy) control (CN) group to a patient (PT) group, capturing the heterogeneity of the underlying disease effect while avoiding confounding variations. More specifically, Smile-GAN learns one mapping function which transform data from CN domain, X, to PT domain, Y, with transformation directions specified by a subtype variable z. An inverse function from Y to Z further regularize the mapping function and can utilized to cluster patients both inside/outside of the training set. Application of the Smile-GAN method on 2832 CN/MCI/Dementia participants identified 4 neurodegenerative patterns/axes which describe two progression pathways. Each pattern has specific implication of cognitive performances and offers better, yet complementary performance in predicting clinical progression compared to measures of amyloid/tau status. Further, by virtue of its generality, Smile-GAN can have broader application beyond Alzheimer’s Disease, and contribute to precision diagnostics and targeted clinical trial recruitment in the broader clinical neurosciences.
