Statistical methods for quantification of voxel-level image noise in PET/CT imaging

Simple item page
dc.check.chapterOfThesisChapter 4en
dc.check.date2028-09-30
dc.check.infoControlled Access
dc.contributor.advisorHuang, Jian
dc.contributor.advisorO'Sullivan, Finbarr
dc.contributor.authorRen, Ranen
dc.contributor.funderScience Foundation Irelanden
dc.date.accessioned2025-05-15T15:12:19Z
dc.date.available2025-05-15T15:12:19Z
dc.date.issued2024
dc.date.submitted2024
dc.descriptionControlled Access
dc.description.abstractPurpose: Clinical positron emission tomography/computed tomography (PET/CT) scan is an important diagnostic tool in modern medicine, e.g. for staging or treatment planning in the field of oncology. Accurate and validated methods for estimating PET image noise are helpful for interpretation of the image and assessment of uncertainty in biomarkers associated with user-defined tissue regions of interest. However, estimation of voxel-level noise characteristics such as variance/covariance is very challenging in clinical PET settings, due to the absence of genuine replicates. With dynamic multi-frame PET image data, it has been possible to construct a practical image-domain bootstrap scheme for estimation of the image noise characteristics. This thesis aims to develop an adaptation of this technique that can be used when only a single data frame, such as the uptake scan, together with the CT attenuation information is retained from the PET study. Method: The approach is based on a Gamma-model representation of the stochastic character of voxel-level uptake data. First, we develop a novel two-step method to fit this model using CT attenuation information and local homogeneity. Step 1: For each voxel, we obtain the initial estimates of the Gamma model parameters by fitting the Gamma distribution to voxel values in a volume of interest (VOI) centered on that voxel. Step 2: We refine the estimates obtained in Step 1 based on simulated effects of attenuation and sensitive correction on noise level. Secondly, we suggest a practical method to estimate the spatial autocorrelation function of an individual clinical PET image. We normalize the image data using the estimated Gamma model parameters and then estimate the spatial autocorrelation function of PET image data by fitting the 3-D auto-regressive model to the normalized data. The proposed estimation methods are verified by extensive experimental studies, including numeric simulation, uniform physical phantom studies, and re-binning lung cancer image data. We found the similarity between the spatial autocorrelation estimated by normalizing single PET image data and the one of uniform physical phantom data scanned with the same scanner. Finally, based on the results of these experimental studies, we proposed two bootstrapping methods to create simulated pseudo-replicates of patient PET data, of which uncertainty is assessed. One is based on a combination of the Gamma model parameters estimation and spatial autocorrelation estimated from uniform physical phantom data scanned with the same scanner. The other is based on a combination of the Gamma model parameters estimation and spatial autocorrelation estimated from normalized single PET image data. Bootstrapping methods are demonstrated in a set of lung cancer patient data collected in a local hospital. Results and conclusions: Estimation of the voxel-level Gamma model parameters is demonstrated on numeric simulation, physical phantom studies, and re-binning lung cancer image data. We first show that the proposed technique can provide a good estimation of the voxel-level Gamma parameters using numerical simulations. Secondly, we show that the proposed technique can provide a good estimation of voxel-level Gamma model parameters and spatial autocorrelation of the uniform physical phantom data collected from a local hospital. Finally, the proposed technique is demonstrated on “replicates” generated via clinical PET data re-binning procedure. The results demonstrate that the suggested method can yield voxel-level Gamma model parameters and spatial autocorrelation estimates very close to those estimated based on "replicates". These results provide empirical support for the proposed bootstrapping technique. Practical use of the bootstrapping method is demonstrated in lung cancer patient data to analyze the standard deviations of standardized uptake values (SUVs) of the malignant regions of interest. Unlike population based test-retest analysis, this methodology focused on patient-specific error and has potential to enhance quantitative decision-making for the patient.en
dc.description.statusNot peer revieweden
dc.description.versionAccepted Versionen
dc.format.mimetypeapplication/pdfen
dc.identifier.citationRen, R. 2024. Statistical methods for quantification of voxel-level image noise in PET/CT imaging. PhD Thesis, University College Cork.
dc.identifier.endpage126
dc.identifier.urihttps://hdl.handle.net/10468/17508
dc.language.isoenen
dc.publisherUniversity College Corken
dc.rights© 2024, Ran Ren.
dc.rights.urihttps://creativecommons.org/licenses/by-nc-nd/4.0/
dc.subjectPET/CTen
dc.subjectRe-binneden
dc.subjectAttenuationen
dc.subjectGamma modelen
dc.subjectBootstrappingen
dc.subjectSpatial autocorrelationen
dc.titleStatistical methods for quantification of voxel-level image noise in PET/CT imaging
dc.typeDoctoral thesisen
dc.type.qualificationlevelDoctoralen
dc.type.qualificationnamePhD - Doctor of Philosophyen
Files
Original bundle
Now showing 1 - 4 of 4
Thumbnail Image
Name:
RanR_PhD2024.pdf
Size:
17.02 MB
Format:
Adobe Portable Document Format
Description:
Full Text E-thesis
Thumbnail Image
Name:
RanR_PhD2024_SubmissionforExaminationForm.pdf
Size:
351.22 KB
Format:
Adobe Portable Document Format
Thumbnail Image
Name:
RanR_PhD2024.tex
Size:
224 KB
Format:
Tex/LateX document
Description:
Tex/LateX document
Thumbnail Image
Name:
RanR_PhD2024 _partial.pdf
Size:
12.99 MB
Format:
Adobe Portable Document Format
Description:
Partial Restriction
License bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
license.txt
Size:
5.2 KB
Format:
Item-specific license agreed upon to submission
Description:
Download
Collections
Research Theses
College of Science, Engineering and Food Science - Doctoral Theses
Mathematical Sciences - Doctoral Theses