Hierarchical Bayesian myocardial perfusion quantification Cian M Scannell

Hierarchical Bayesian myocardial perfusion quantification Cian M. Scannell 1, 2, Amedeo Chiribiri 1, Adriana D. M. Villa 1, Marcel Breeuwer 3, 4, Jack Lee 1 1 King’s College London, United Kingdom 2 The Alan Turing Institute, United Kingdom 3 Philips Healthcare, The Netherlands 4 Eindhoven University of Technology, The Netherlands

Declaration of Financial Interests or Relationships Speaker Name: Cian M. Scannell I have the following financial interest or relationship(s) to disclose with regard to the subject matter of this presentation: • Grant/research support: Philips Healthcare

Myocardial perfusion quantification Arterial Input Function Myocardial Tissue Curve 3/13

Myocardial perfusion quantification Arterial Input Function 2 CXM Myocardial Tissue Curve 4/13

The reliability of the parameter estimates but the issue of parameter uniqueness is further complicated by the surfeit of possible parameter combinations. UNCERTAINTY IN THE ANALYSIS OF TRACER KINETICS USING DYNAMIC CONTRAST-ENHANCED T 1 -WEIGHTED MRI -- DAVID L. BUCKLEY, MAGNETIC RESONANCE IN MEDICINE, 2002 5/13

The reliability of the parameter estimates A shortcoming with complex pharmacokinetic models is that identical tissue curves can be generated using a single arterial input function and multiple sets of perfusion model parameters ESTIMATING EXTRACTION FRACTION AND BLOOD FLOW BY COMBINING FIRSTPASS MYOCARDIAL PERFUSION AND T 1 MAPPING RESULTS -- LIKHITE ET AL. , QUANTITATIVE IMAGING IN MEDICINE AND SURGERY, 2017 1. Schwab et al. , MRM, 2015 2. Broadbent et al. , MRM, 2013 3. Jerosch-Herold et al. , Int. J. Card. Imaging, 1999 6/13

The reliability of the parameter estimates 7/13

The solution – Bayesian inference Bayesian Inference in MRI, Member Initiated Symposium ISMRM 2019 “Target Audience: Scientists and clinicians interested in extracting meaningful information from noisy data. ” Least squares fitting: Sum of squared errors Likelihood function Bayesian inference: 8/13

Choice of priors – hierarchical Bayes 2 1 Avoids using fixed prior distributions 9/13

Results – simulation study NMSE Parameter Bayesian Least squares p-value All 0. 13 (0. 2) 0. 32 (0. 55) p < 0. 0001 0. 05 (0. 09) 0. 1 (0. 09) p = 0. 002 0. 22 (0. 27) 0. 35 (0. 31) p = 0. 02 0. 12 (0. 16) 0. 20 (0. 17) p = 0. 01 0. 11 (0. 21) 0. 63 (0. 96) p < 0. 0001 PS 10/13