Optimizability of loglikelihoods for the estimation of detector efficiencies and singles rates in PET

Matthew W. Jacobson, Kris Thielemans

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In various prior work, Poisson models have been proposed for normalization scan and delayed coincidence measurements in PET systems. In these models, the mean measured counts are given by pair-wise products of a set of unknown detector efficiency or singles rate parameters. Additionally, several kinds of iterative algorithms have been proposed for solving the associated maximum likelihood estimation problem, and have exhibited strong empirical performance. To our knowledge, however, no analysis has been given proving that these algorithms actually converge, but only that convergence, should it hypothetically occur, would have to be to a stationary point of the loglikelihood. One of the main obstacles to a complete convergence analysis is showing that the loglikelihood actually possesses a finite maximizer. As we will show, there are cases where the loglikelihood function for these applications can approach its maximum value only by letting the unknown parameters tend asymptotically to infinity. In such cases, convergent behavior is not possible with any algorithm because there is nothing to converge to. In this article, we derive sufficient conditions - for cylindrical PET system geometries - guaranteeing not only that a unique and finite maximizer exists, but also that the loglikelihood maximization is equivalent to minimizing a strictly convex function. The latter obviates many questions of convergent algorithm design, since virtually any minimization algorithm with standard monotonicity and stationary properties will convergently minimize a strictly convex function when the minimizer exists. The conditions that we derive are, oreover, readily verifiable by preanalysis of the projection measurements and show that failure of a solution to exist is a very rare event, xcept for unrealistically low count scans or for systems with unrealistically dense patterns of malfunctioning detectors.

Original languageEnglish (US)
Title of host publication2008 IEEE Nuclear Science Symposium Conference Record, NSS/MIC 2008
Pages4580-4586
Number of pages7
DOIs
StatePublished - 2008
Externally publishedYes
Event2008 IEEE Nuclear Science Symposium Conference Record, NSS/MIC 2008 - Dresden, Germany
Duration: Oct 19 2008Oct 25 2008

Other

Other2008 IEEE Nuclear Science Symposium Conference Record, NSS/MIC 2008
CountryGermany
CityDresden
Period10/19/0810/25/08

Fingerprint

detectors
infinity
projection
optimization
products
geometry

Keywords

  • Detector efficiencies
  • Maximum likelihood
  • PET
  • Singles

ASJC Scopus subject areas

  • Radiation
  • Nuclear and High Energy Physics
  • Radiology Nuclear Medicine and imaging

Cite this

Jacobson, M. W., & Thielemans, K. (2008). Optimizability of loglikelihoods for the estimation of detector efficiencies and singles rates in PET. In 2008 IEEE Nuclear Science Symposium Conference Record, NSS/MIC 2008 (pp. 4580-4586). [4774352] https://doi.org/10.1109/NSSMIC.2008.4774352

Optimizability of loglikelihoods for the estimation of detector efficiencies and singles rates in PET. / Jacobson, Matthew W.; Thielemans, Kris.

2008 IEEE Nuclear Science Symposium Conference Record, NSS/MIC 2008. 2008. p. 4580-4586 4774352.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Jacobson, MW & Thielemans, K 2008, Optimizability of loglikelihoods for the estimation of detector efficiencies and singles rates in PET. in 2008 IEEE Nuclear Science Symposium Conference Record, NSS/MIC 2008., 4774352, pp. 4580-4586, 2008 IEEE Nuclear Science Symposium Conference Record, NSS/MIC 2008, Dresden, Germany, 10/19/08. https://doi.org/10.1109/NSSMIC.2008.4774352
Jacobson MW, Thielemans K. Optimizability of loglikelihoods for the estimation of detector efficiencies and singles rates in PET. In 2008 IEEE Nuclear Science Symposium Conference Record, NSS/MIC 2008. 2008. p. 4580-4586. 4774352 https://doi.org/10.1109/NSSMIC.2008.4774352
Jacobson, Matthew W. ; Thielemans, Kris. / Optimizability of loglikelihoods for the estimation of detector efficiencies and singles rates in PET. 2008 IEEE Nuclear Science Symposium Conference Record, NSS/MIC 2008. 2008. pp. 4580-4586
@inproceedings{30a5a435ea224806beff4c4a74af16e0,
title = "Optimizability of loglikelihoods for the estimation of detector efficiencies and singles rates in PET",
abstract = "In various prior work, Poisson models have been proposed for normalization scan and delayed coincidence measurements in PET systems. In these models, the mean measured counts are given by pair-wise products of a set of unknown detector efficiency or singles rate parameters. Additionally, several kinds of iterative algorithms have been proposed for solving the associated maximum likelihood estimation problem, and have exhibited strong empirical performance. To our knowledge, however, no analysis has been given proving that these algorithms actually converge, but only that convergence, should it hypothetically occur, would have to be to a stationary point of the loglikelihood. One of the main obstacles to a complete convergence analysis is showing that the loglikelihood actually possesses a finite maximizer. As we will show, there are cases where the loglikelihood function for these applications can approach its maximum value only by letting the unknown parameters tend asymptotically to infinity. In such cases, convergent behavior is not possible with any algorithm because there is nothing to converge to. In this article, we derive sufficient conditions - for cylindrical PET system geometries - guaranteeing not only that a unique and finite maximizer exists, but also that the loglikelihood maximization is equivalent to minimizing a strictly convex function. The latter obviates many questions of convergent algorithm design, since virtually any minimization algorithm with standard monotonicity and stationary properties will convergently minimize a strictly convex function when the minimizer exists. The conditions that we derive are, oreover, readily verifiable by preanalysis of the projection measurements and show that failure of a solution to exist is a very rare event, xcept for unrealistically low count scans or for systems with unrealistically dense patterns of malfunctioning detectors.",
keywords = "Detector efficiencies, Maximum likelihood, PET, Singles",
author = "Jacobson, {Matthew W.} and Kris Thielemans",
year = "2008",
doi = "10.1109/NSSMIC.2008.4774352",
language = "English (US)",
isbn = "9781424427154",
pages = "4580--4586",
booktitle = "2008 IEEE Nuclear Science Symposium Conference Record, NSS/MIC 2008",

}

TY - GEN

T1 - Optimizability of loglikelihoods for the estimation of detector efficiencies and singles rates in PET

AU - Jacobson, Matthew W.

AU - Thielemans, Kris

PY - 2008

Y1 - 2008

N2 - In various prior work, Poisson models have been proposed for normalization scan and delayed coincidence measurements in PET systems. In these models, the mean measured counts are given by pair-wise products of a set of unknown detector efficiency or singles rate parameters. Additionally, several kinds of iterative algorithms have been proposed for solving the associated maximum likelihood estimation problem, and have exhibited strong empirical performance. To our knowledge, however, no analysis has been given proving that these algorithms actually converge, but only that convergence, should it hypothetically occur, would have to be to a stationary point of the loglikelihood. One of the main obstacles to a complete convergence analysis is showing that the loglikelihood actually possesses a finite maximizer. As we will show, there are cases where the loglikelihood function for these applications can approach its maximum value only by letting the unknown parameters tend asymptotically to infinity. In such cases, convergent behavior is not possible with any algorithm because there is nothing to converge to. In this article, we derive sufficient conditions - for cylindrical PET system geometries - guaranteeing not only that a unique and finite maximizer exists, but also that the loglikelihood maximization is equivalent to minimizing a strictly convex function. The latter obviates many questions of convergent algorithm design, since virtually any minimization algorithm with standard monotonicity and stationary properties will convergently minimize a strictly convex function when the minimizer exists. The conditions that we derive are, oreover, readily verifiable by preanalysis of the projection measurements and show that failure of a solution to exist is a very rare event, xcept for unrealistically low count scans or for systems with unrealistically dense patterns of malfunctioning detectors.

AB - In various prior work, Poisson models have been proposed for normalization scan and delayed coincidence measurements in PET systems. In these models, the mean measured counts are given by pair-wise products of a set of unknown detector efficiency or singles rate parameters. Additionally, several kinds of iterative algorithms have been proposed for solving the associated maximum likelihood estimation problem, and have exhibited strong empirical performance. To our knowledge, however, no analysis has been given proving that these algorithms actually converge, but only that convergence, should it hypothetically occur, would have to be to a stationary point of the loglikelihood. One of the main obstacles to a complete convergence analysis is showing that the loglikelihood actually possesses a finite maximizer. As we will show, there are cases where the loglikelihood function for these applications can approach its maximum value only by letting the unknown parameters tend asymptotically to infinity. In such cases, convergent behavior is not possible with any algorithm because there is nothing to converge to. In this article, we derive sufficient conditions - for cylindrical PET system geometries - guaranteeing not only that a unique and finite maximizer exists, but also that the loglikelihood maximization is equivalent to minimizing a strictly convex function. The latter obviates many questions of convergent algorithm design, since virtually any minimization algorithm with standard monotonicity and stationary properties will convergently minimize a strictly convex function when the minimizer exists. The conditions that we derive are, oreover, readily verifiable by preanalysis of the projection measurements and show that failure of a solution to exist is a very rare event, xcept for unrealistically low count scans or for systems with unrealistically dense patterns of malfunctioning detectors.

KW - Detector efficiencies

KW - Maximum likelihood

KW - PET

KW - Singles

UR - http://www.scopus.com/inward/record.url?scp=67649194616&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=67649194616&partnerID=8YFLogxK

U2 - 10.1109/NSSMIC.2008.4774352

DO - 10.1109/NSSMIC.2008.4774352

M3 - Conference contribution

AN - SCOPUS:67649194616

SN - 9781424427154

SP - 4580

EP - 4586

BT - 2008 IEEE Nuclear Science Symposium Conference Record, NSS/MIC 2008

ER -