Show simple item record

dc.contributor.authorWalker, M D
dc.contributor.authorAsselin, M-C
dc.contributor.authorJulyan, Peter J
dc.contributor.authorFeldmann, M
dc.contributor.authorTalbot, P S
dc.contributor.authorJones, T
dc.contributor.authorMatthews, J C
dc.date.accessioned2012-01-09T22:47:31Z
dc.date.accessioned2012-01-09T22:47:33Z
dc.date.available2012-01-09T22:47:31Z
dc.date.available2012-01-09T22:47:33Z
dc.date.issued2011-02-21
dc.identifier.citationBias in iterative reconstruction of low-statistics PET data: benefits of a resolution model. 2011, 56 (4):931-49 Phys Med Biolen
dc.identifier.issn1361-6560
dc.identifier.pmid21248391
dc.identifier.doi10.1088/0031-9155/56/4/004
dc.identifier.urihttp://hdl.handle.net/10541/201092
dc.identifier.urihttp://hdl.handle.net/10541/201087
dc.description.abstractIterative image reconstruction methods such as ordered-subset expectation maximization (OSEM) are widely used in PET. Reconstructions via OSEM are however reported to be biased for low-count data. We investigated this and considered the impact for dynamic PET. Patient listmode data were acquired in [(11)C]DASB and [(15)O]H(2)O scans on the HRRT brain PET scanner. These data were subsampled to create many independent, low-count replicates. The data were reconstructed and the images from low-count data were compared to the high-count originals (from the same reconstruction method). This comparison enabled low-statistics bias to be calculated for the given reconstruction, as a function of the noise-equivalent counts (NEC). Two iterative reconstruction methods were tested, one with and one without an image-based resolution model (RM). Significant bias was observed when reconstructing data of low statistical quality, for both subsampled human and simulated data. For human data, this bias was substantially reduced by including a RM. For [(11)C]DASB the low-statistics bias in the caudate head at 1.7 M NEC (approx. 30 s) was -5.5% and -13% with and without RM, respectively. We predicted biases in the binding potential of -4% and -10%. For quantification of cerebral blood flow for the whole-brain grey- or white-matter, using [(15)O]H(2)O and the PET autoradiographic method, a low-statistics bias of <2.5% and <4% was predicted for reconstruction with and without the RM. The use of a resolution model reduces low-statistics bias and can hence be beneficial for quantitative dynamic PET.
dc.language.isoenen
dc.subject.meshBenzylamines
dc.subject.meshCarbon Radioisotopes
dc.subject.meshHumans
dc.subject.meshImage Processing, Computer-Assisted
dc.subject.meshKinetics
dc.subject.meshModels, Biological
dc.subject.meshMonte Carlo Method
dc.subject.meshOxygen Radioisotopes
dc.subject.meshPositron-Emission Tomography
dc.subject.meshWater
dc.titleBias in iterative reconstruction of low-statistics PET data: benefits of a resolution model.en
dc.typeArticleen
dc.contributor.departmentSchool of Cancer and Enabling Sciences, Wolfson Molecular Imaging Centre, MAHSC, The University of Manchester, Manchester, UK.en
dc.identifier.journalPhysics in Medicine and Biologyen
html.description.abstractIterative image reconstruction methods such as ordered-subset expectation maximization (OSEM) are widely used in PET. Reconstructions via OSEM are however reported to be biased for low-count data. We investigated this and considered the impact for dynamic PET. Patient listmode data were acquired in [(11)C]DASB and [(15)O]H(2)O scans on the HRRT brain PET scanner. These data were subsampled to create many independent, low-count replicates. The data were reconstructed and the images from low-count data were compared to the high-count originals (from the same reconstruction method). This comparison enabled low-statistics bias to be calculated for the given reconstruction, as a function of the noise-equivalent counts (NEC). Two iterative reconstruction methods were tested, one with and one without an image-based resolution model (RM). Significant bias was observed when reconstructing data of low statistical quality, for both subsampled human and simulated data. For human data, this bias was substantially reduced by including a RM. For [(11)C]DASB the low-statistics bias in the caudate head at 1.7 M NEC (approx. 30 s) was -5.5% and -13% with and without RM, respectively. We predicted biases in the binding potential of -4% and -10%. For quantification of cerebral blood flow for the whole-brain grey- or white-matter, using [(15)O]H(2)O and the PET autoradiographic method, a low-statistics bias of <2.5% and <4% was predicted for reconstruction with and without the RM. The use of a resolution model reduces low-statistics bias and can hence be beneficial for quantitative dynamic PET.


This item appears in the following Collection(s)

Show simple item record