A FORMAL FRAMEWORK TO CONSTRUCT SIMPLIFIED MODELS OF PROBABILISTIC DESCRIPTIONS OF NEUROANATOMICAL DATA

DIAZ CELAURO, LUCAS; ARENAZA, BAUTISTA; VALLEJO AZAR, MARIANA; ELIZALDE-ACEVEDO, BAUTISTA; ALBA FERRARA, L.; BENDERSKI, MARIANA; GONZALEZ, PAULA; SAMENGO, INES

santa fe

Congreso; congreso MACI; 2023

maci

Abstract:In physics, an effective description of a macroscopic can sometimes be derived from the microscopic variables andthe fundamental laws. This procedure is often referred to as “coarse graining”, because the detailed state of the micro-scopic system is abandoned in favour of a model in which some emergent variables and some effective laws controlthe dynamics. A crucial step is to decide which variables to preserve and which to discard, so as to obtain the simplestdescription that attains minimal information loss. Machta and co-workers have derived a principled strategy, in whichthe preserved variables are defined by the eigenvectors of the Fisher information with maximal eigenvalues. We hereapply this method to simplify the high-dimensional probability distribution describing the population variability of thegeometrical properties of the cerebral cortex, as obtained from MRI images of 193 healthy volunteers. The simplifiedmodel reduces the number of parameters by a factor of 3, while still providing a good approximation of the originaldistribution. The performance of the reduction is assessed by several validation tests and the simplified descriptionhas significant implications for the understanding of the brain’s anatomical properties.