Logo of the Physikalisch-Technische Bundesanstalt

Bayesian neural network to quantify perfusion parameters and their uncertainties


Physiological parameter estimation by MRI is affected by intrinsic ambiguity in the data such as noise or model inaccuracies. Edengenet Dejene (PTB 8.14), together with Winfried Brenner (Charité), Marcus R. Makowski (TU München), and Christoph Kolbitsch (PTB 8.14), now developed a method for the uncertainty quantification of physiological parameters in dynamic contrast enhanced MRI of the liver. They use a Bayesian neural network framework for the accurate estimation of perfusion parameters of this organ and are able to quantify the aleatoric and epistemic uncertainties associated with these parameters on a pixel-by pixel basis. This is important for clinical applications, e.g. hepatic tumor characterization, because it could indicate cases where the trained model is inadequate and additional training with an adapted training data set is required.

This work has now been published in Physics in Medicine and Biology veröffentlicht (doi: 10.1088/1361-6560/ad0284).


Edengenet Dejene, E-Mail: 📧 edengenet.dejene(at)ptb.de