[
home]
[
Personal Program]
[
Help]
tag
16:00
0 mins
Predicting Uncertainty of Metabolite Quantification in Magnetic Resonance Spectroscopy with Applications for Adaptive Ensembling
Julian Merkofer, Sina Amirrajab, Johan van den Brink, Mitko Veta, Jacobus Jansen, Marcel Breeuwer, Ruud van Sloun
Session: Poster Session 1 (Even numbers)
Session starts: Thursday 26 January, 16:00
Presentation starts: 16:00
Julian Merkofer (Eindhoven University of Technology)
Sina Amirrajab (Eindhoven University of Technology)
Johan van den Brink (Philips Healthcare)
Mitko Veta (Eindhoven University of Technology)
Jacobus Jansen (Maastricht University Medical Center+)
Marcel Breeuwer (Eindhoven University of Technology, Philips Healthcare)
Ruud van Sloun (Eindhoven University of Technology, Philips Research)
Abstract:
Current deep learning methods for metabolite quantification in MR spectroscopy do not offer reliable measures for uncertainty. Having such a measure would not only flag potential (self-identified) fitting errors, but also enable uncertainty-based adaptive ensembling of classic model-based fitting and deep learning predictions. In this abstract, we propose a training strategy based on a log-likelihood cost that allows joint optimization of metabolite concentrations and uncertainty estimation for each individual metabolite. On synthetic data, we show that the predicted uncertainty correlates well with the actual estimation error and that uncertainty-based adaptive ensembling outperforms the individual estimators as well as standard ensembling.