TY - GEN
T1 - Optimization of image reconstruction for the RatCAP (PET) tomograph
T2 - 2007 IEEE Nuclear Science Symposium and Medical Imaging Conference, NSS-MIC
AU - Southekal, Sudeepti S.
AU - Purschke, Martin
AU - Schlyer, David J.
AU - Woody, Craig L.
AU - Vaska, Paul
PY - 2007
Y1 - 2007
N2 - A highly accurate system model is the basis of statistical iterative reconstruction for the RatCAP. The model is used to generate a fully 3D, Monte Carlo system response matrix (SRM) for Maximum Likelihood Expectation Maximization (MLEM) reconstruction. Significant efforts have been taken to ensure a faithful match to the actual tomograph. One of the main considerations with Monte Carlo SRMs is statistical accuracy, as any error in the matrix could propagate into the reconstructed image. In theory, it is possible to simulate an arbitrarily large number of events, making the statistical errors in the matrix insignificant compared to the data. However, at a certain point, errors in the data limit any further improvement in accuracy due to higher statistics in the matrix. An effort to achieve the best possible quantitative accuracy, while optimizing the tradeoff between model accuracy and computation time is presented. Realistic rat brain simulations have been reconstructed using multiple realizations of system matrices at varying count levels. The sensitivity of our methods to the errors in the data, as well as the reconstruction algorithm has been analyzed. The overall goal of this study is to find the SRM with the best tradeoff between resolution and noise for our reconstruction, and simultaneously validate the use of our model for quantitative analyses with the RatCAP.
AB - A highly accurate system model is the basis of statistical iterative reconstruction for the RatCAP. The model is used to generate a fully 3D, Monte Carlo system response matrix (SRM) for Maximum Likelihood Expectation Maximization (MLEM) reconstruction. Significant efforts have been taken to ensure a faithful match to the actual tomograph. One of the main considerations with Monte Carlo SRMs is statistical accuracy, as any error in the matrix could propagate into the reconstructed image. In theory, it is possible to simulate an arbitrarily large number of events, making the statistical errors in the matrix insignificant compared to the data. However, at a certain point, errors in the data limit any further improvement in accuracy due to higher statistics in the matrix. An effort to achieve the best possible quantitative accuracy, while optimizing the tradeoff between model accuracy and computation time is presented. Realistic rat brain simulations have been reconstructed using multiple realizations of system matrices at varying count levels. The sensitivity of our methods to the errors in the data, as well as the reconstruction algorithm has been analyzed. The overall goal of this study is to find the SRM with the best tradeoff between resolution and noise for our reconstruction, and simultaneously validate the use of our model for quantitative analyses with the RatCAP.
UR - https://www.scopus.com/pages/publications/48149097208
U2 - 10.1109/NSSMIC.2007.4436774
DO - 10.1109/NSSMIC.2007.4436774
M3 - Conference contribution
AN - SCOPUS:48149097208
SN - 1424409233
SN - 9781424409235
T3 - IEEE Nuclear Science Symposium Conference Record
SP - 3051
EP - 3054
BT - 2007 IEEE Nuclear Science Symposium and Medical Imaging Conference, NSS-MIC
Y2 - 27 October 2007 through 3 November 2007
ER -