Quantification of functional imaging biomarkers in medicine: technical validation and simplification

Thomas Koopman

    Research output: PhD ThesisPhD-Thesis - Research and graduation internal

    127 Downloads (Pure)

    Abstract

    This thesis describes various technical aspects of validation of quantitative imaging biomarkers derived from different imaging studies (PET, PET-CT, MRI) for various targets (perfusion, diffusion, cell metabolism). For the quantification of each of these biomarkers a combination of system modelling, signal processing and parameter estimation is required. Several aspects hereof (such as model selection, fitting routines and signal extraction methods) have been investigated in this PhD project. A recurring theme in these studies is the accuracy and precision of a quantitative imaging biomarker. Precision or repeatability can be assessed using test-retest studies. With poor precision, the clinical value of a biomarker is limited. Moreover, clinical applicability of a method is important for implementation in routine care. The research described in this thesis therefore focused on optimization as well as simplification of quantification methods. The study in chapter two investigated the possibility of simplifying the clinical protocol in tracer kinetic modelling of brain perfusion with [15O]H2O PET. The study looked into the possibility of calculating perfusion without knowing the arterial input function. The study confirmed that without the AIF relative quantification is still possible. Although none of the methods was able to provide an accurate estimate of absolute perfusion, one method provided a reasonable precision and could therefore be used to study longitudinal changes. The aim of the study in chapter three was, firstly, to derive the optimal plasma input kinetic model for dynamic O-(2-[18F]fluoroethyl)-L-tyrosine ([18F]FET) PET studies and, secondly, to use that model as a reference to evaluate simplified methods. [18F]FET is a PET tracer used in cancer imaging, in particular for brain tumours. The tracer has been shown to be very sensitive in detecting neoplasia in the brain and is therefore useful in determining the tumour extent in glioma. The optimal model was the reversible two-tissue compartment model. In chapter four, the data from chapter three was processed at the voxel level (as opposed to region level). This is useful for evaluation of tracer uptake distribution within a tumour or to be able to delineate the tumour extent. (Non-invasive) Logan graphical analysis provided volume of distribution (ratio) maps with the lowest level of noise, but poor accuracy, while the basis function implementations provided the best accuracy, but also high noise levels. SUV ratio maps provided better results if later interval times were used. Chapter five investigated the precision of image-derived arterial input functions obtained with dynamic contrast enhanced MRI in head and neck cancer patients. The arterial input function (AIF) is necessary to estimate pharmacokinetic parameters with dynamic contrast enhanced MRI. The results show that accurate measurement of an image-derived AIF is unlikely in the head and neck region. Usage of the population averaged AIF is therefore recommended. The intravoxel incoherent motion (IVIM) model for diffusion-weighted imaging may provide useful biomarkers for disease management in head and neck cancer. In chapter six, the Bayesian and neural network approaches substantially outperformed conventional nonlinear regression in terms of test-retest repeatability. The processing speed of the neural network makes it viable for use in clinical practice. However, the approach needs to be further improved to identify neural networks that are both consistent and precise. Finally, chapter seven shows that [18F]FDG PET radiomics provides additional prognostic value when combined with clinical information and first-order [18F]FDG PET imaging biomarkers. The results show that latent factors can improve prediction of recurrence, distant metastasis and overall survival. Moreover, the study shows how this information can be used for personalized risk-stratification of patients’ outcome. Better prognosis prediction in locally advanced head and neck squamous cell carcinoma can thus optimize personalized cancer care.
    Original languageEnglish
    QualificationPhD
    Awarding Institution
    • Vrije Universiteit Amsterdam
    Supervisors/Advisors
    • Boellaard, Ronald, Supervisor, -
    • Castelijns, J.A., Supervisor, -
    • Yaqub, Mohammed Maqsood, Co-supervisor, -
    • Marcus, J.T., Co-supervisor
    Award date8 Apr 2022
    Place of Publications.l.
    Publisher
    Publication statusPublished - 8 Apr 2022

    Keywords

    • Medical Image Processing
    • Biomarkers
    • Positron-Emission Tomography
    • Magnetic Resonance Imaging
    • Perfusion Imaging
    • Diffusion Magnetic Resonance Imaging
    • Data Accuracy
    • Pharmacokinetic Modeling

    Fingerprint

    Dive into the research topics of 'Quantification of functional imaging biomarkers in medicine: technical validation and simplification'. Together they form a unique fingerprint.

    Cite this