Talk: Machine Learning and Mathematical Methods for the Brain Age Problem
Speaker: Alessandro Maria Selvitella, Purdue University Fort Wayne (grid.503846.c)
Title: Machine Learning and Mathematical Methods for the Brain Age Problem
Emcee: Cheng Xue
Backend host: Gelana Tostaeva
Details: https://neuromatch.io/abstract?submis...
Lab website: https://users.pfw.edu/aselvite/
Paper link: https://doi.org/10.7771/2158-4052.1445
Presented during Neuromatch Conference 3.0, Oct 26-30, 2020.
Summary: Overall, the global population is aging and, as a result, there is an increasing frequency of age-related diseases. Researchers have tried to establish neuroanatomical biomarkers of aging and understand how the healthy adult brain changes with age. This is the Brain Age Problem (BAP). Our research concentrates on the use of brain biometrics, machine learning methods, and pure mathematics to improve the current answers to the Brain Age Problem (BAP). In this work, we derived the Normalized Stretching Ratio (NSR) given by (Thickness)^2/(Surface Area), which is a surrogate of the isoperimetric ratio that quantifies the stretching of a brain region. We used the NSRs of 68 brain regions as the predictors of a classification model on a freely available data set of young adults, collected by the Human Connectome Project. Our model does not sacrifice prediction accuracy when compared to a similar model on the raw data and so the dimensionality reduction provided by the NSRs yields a simpler explanation of the changes that occur in the brain. Our model reduces the gap between explainability and prediction accuracy by merging ideas from pure mathematical fields, such as differential geometry, machine learning, and neuroscience. This is joint work with my students Justin Asher, Khoa Tan Dang, Peter Klopfenstein, Maxwell Masters, and Jucoen Yeater.