Title: Machine learning in neuroimaging: A journey from mass-univariate methods to multi-variate imaging signatures of diseases, in search of precision diagnostics
Abstract: In this talk, I will briefly provide a historical perspective of the past 20 years, in which we saw the transition from hypothesis-based to big-data-driven methods of analysis of brain images. This journey has been motivated largely by the need for more specific and personalized imaging signatures of diseases and their subtypes, which will eventually open the door for early detection and personalized prognosis and treatment. I present examples from studies of aging, Alzheimer’s Disease, brain development, schizophrenia, and brain cancer. Moreover, I discuss challenges in machine learning emanating from multi-site settings, disease heterogeneity, and the need for interpretability and derivation of statistical significance maps in multivariate settings. I conclude with work in our laboratory toward addressing these challenges.