TOFMOD Workshop 2020 (satellite event of Virtual Physiological Human conference 2020)

Speaker: Katerina Skardova, Department of Mathematics, Faculty of Nuclear Sciences and Physical Engineering, Czech Technical University in Prague, Czech Republic

Title: Neural networks and mathematical models

Abstract: In this contribution, we discuss how numerical simulations, machine learning based on neural networks, and experimental data could be combined in order to create an efficient framework for physiological parameter estimation. We deal with the estimation of T1 relaxation time based on image series acquired by the Modified Look-Locker Inversion Recovery (MOLLI) magnetic resonance imaging sequence.  Before the pixel-wise T1 relaxation time estimation, the frames of the image sequence need to be registered. We present a locally adjusted optical flow-based method [1] for spatial registration of the image sequence. The T1 estimation method itself, combines machine learning and modeling methods.  A simplified mathematical model of the imaging sequence is used to generate a sufficiently large training dataset for the neural network, which then provides the first estimation of the T1 relaxation time. This estimation is then improved by standard optimization techniques. 

[1] Skardova, K., Oberhuber, T., Tintera, J., & Chabiniok, R. (2018). Signed-distance function based non-rigid registration of image series with varying image intensity. Discrete & Continuous Dynamical Systems-S.