TY - GEN
T1 - Application of Markov chain Monte Carlo for uncertainty quantification in quantitative imaging problems
AU - Bledsoe, Keith C.
AU - Jessee, Matthew A.
AU - Blackston, Matthew A.
AU - Knowles, Justin R.
AU - Ziock, Klaus Peter
AU - Lefebvre, Jordan P.
N1 - Publisher Copyright:
© 2019 American Nuclear Society. All rights reserved.
PY - 2019
Y1 - 2019
N2 - The DiffeRential Evolution Adaptive Metropolis (DREAM) method, a Markov chain Monte Carlo approach, is used for optimization and uncertainty analysis of a radioactive source–shield system using pixelated measured data generated by radiation imagers, where each pixel in the image contains a gamma-ray spectrum with statistical uncertainty. DREAM uses this spectral information to determine source thickness and source strength while also propagating uncertainty from the measured data to the solution. Using measurements simulated with the GEANT4 code, successful parameter reconstruction is demonstrated on a numerical test case in a slab-shield geometry. In the presence of statistical noise of 5%–9%, parameters are calculated to better than 95% with 2σ error bars that generally encompass the actual values. In a test problem with real-world measurements, source strength per energy line is calculated to within 78%–97% of the actual value.
AB - The DiffeRential Evolution Adaptive Metropolis (DREAM) method, a Markov chain Monte Carlo approach, is used for optimization and uncertainty analysis of a radioactive source–shield system using pixelated measured data generated by radiation imagers, where each pixel in the image contains a gamma-ray spectrum with statistical uncertainty. DREAM uses this spectral information to determine source thickness and source strength while also propagating uncertainty from the measured data to the solution. Using measurements simulated with the GEANT4 code, successful parameter reconstruction is demonstrated on a numerical test case in a slab-shield geometry. In the presence of statistical noise of 5%–9%, parameters are calculated to better than 95% with 2σ error bars that generally encompass the actual values. In a test problem with real-world measurements, source strength per energy line is calculated to within 78%–97% of the actual value.
KW - Markov chain Monte Carlo
KW - Quantitative imaging
KW - Uncertainty quantification
UR - http://www.scopus.com/inward/record.url?scp=85075338721&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85075338721
T3 - International Conference on Mathematics and Computational Methods Applied to Nuclear Science and Engineering, M and C 2019
SP - 1928
EP - 1937
BT - International Conference on Mathematics and Computational Methods Applied to Nuclear Science and Engineering, M and C 2019
PB - American Nuclear Society
T2 - 2019 International Conference on Mathematics and Computational Methods Applied to Nuclear Science and Engineering, M and C 2019
Y2 - 25 August 2019 through 29 August 2019
ER -