Meeting Banner
Abstract #2623

Non central chi estimation of multi-compartment models improves model selection by reducing overfitting

Aymeric Stamm 1 , Benoit Scherrer 1 , Stefano Baraldo 2 , Olivier Commowick 3 , and Simon Warfield 1

1 Computational Radiology Laboratory, Boston Children's Hospital, Boston, MA, United States, 2 MOX, Politecnico di Milano, Milan, Italy, 3 VISAGES, INRIA, Rennes, France, Metropolitan

Noise in diffusion MRI is known to be characterized by a non-central chi distribution. Many denoising methods have accounted for this but, for the estimation of diffusion models, the noise is most of the time still approximated by a Gaussian distribution. In this abstract, we examine the impact of this approximation to determine the optimal number of fascicles required for the estimation of multi-compartment models. We show that performing the models' estimation within a non-central chi framework significantly reduces over-fitting thus yielding a more reliable selection of the optimal number of fascicles.

This abstract and the presentation materials are available to members only; a login is required.

Join Here