Meeting Banner
Abstract #4239

Estimation of Parameters from Sparsely Sampled in-vivo Magnetization Transfer Data Using Artificial Neural Networks

Henrik Marschner1, Dirk K. Mller1, Andr Pampel1, Jane Neumann1, Harald E. Mller1

1Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany

We examine whether Artificial Neural Networks (ANN) can be trained to estimate MT parameter sets from sparsely sampled in-vivo MT data. ANNs were trained using densely sampled MT data from healthy volunteers and the MT parameters obtained using conventional fitting as input. ANNs were used to extract MT parameters from sparsely sampled data. The obtained parameters were compared with those that come out using the conventional method. It is shown that parameters obtained with both methods are highly correlated (R>0.97). Once ANNs are trained subsequent measurements of other individuals and parameter estimation can be notably accelerated.

Abstract PDF

Keywords

added addition advisable already array artificial assumption assumptions audience bath binary bound boundary brain channel choosing colored configurations conjugate considering consisted consisting constant constants constrained constraints datasets definition demonstrating densely described describing dimensional dirk display displayed distributed enable entire equally error estimation even examine exceeding exchange excluded fast feed field fitted fitting forward fraction frequencies frequency function furthermore general generated gradient gray hardware head healthy hidden highly hyperbolic implementation improved in vivo included indirectly inhomogeneities input inside just learned least linear liquid macromolecules magnetization mandatory maps matching measured model necessary necessity needs network networks neural neurons often optimal particular performance plus pool pools prevent previously processes protons pulse quantification quantitative randomly recently reduction relation remaining required respectively sample sampled saturation scaled schemes sciences selected semi semisolid sets solid sparsely square squares stopping subjects sufficient sufficiently suitable tangent target throughout trained training transfer transverse typically unseen validation various vector visible volunteers whereby white zoom