G5 Doctoral dissertation (article)
Algorithmic Analysis Techniques for Molecular Imaging




List of Authors: Harri Merisaari
Publisher: TUCS Dissertations
Place: Turku
Publication year: 2016
ISBN: 978-952-12-3442-2

Abstract
This study addresses image processing techniques for two medical imaging
modalities: Positron Emission Tomography (PET) and Magnetic Resonance
Imaging (MRI), which can be used in studies of human body functions and
anatomy in a non-invasive manner.
In PET, the so-called Partial Volume Effect (PVE) is caused by low
spatial resolution of the modality. The efficiency of a set of PVE-correction
methods is evaluated in the present study. These methods use information
about tissue borders which have been acquired with the MRI technique. As
another technique, a novel method is proposed for MRI brain image segmen-
tation.
A standard way of brain MRI is to use spatial prior information
in image segmentation. While this works for adults and healthy neonates,
the large variations in premature infants preclude its direct application.
The proposed technique can be applied to both healthy and non-healthy
premature infant brain MR images. Diffusion Weighted Imaging (DWI) is
a MRI-based technique that can be used to create images for measuring
physiological properties of cells on the structural level.
We optimise the
scanning parameters of DWI so that the required acquisition time can be
reduced while still maintaining good image quality.
In the present work, PVE correction methods, and physiological DWI
models are evaluated in terms of repeatabilityof the results. This gives in-
formation on the reliability of the measures given by the methods. The
evaluations are done using physical phantom objects, correlation measure-
ments against expert segmentations, computer simulations with realistic
noise modelling, and with repeated measurements conducted on real pa-
tients. In PET, the applicability and selection of a suitable partial volume
correction method was found to depend on the target application. For MRI,
the data-driven segmentation offers an alternative when using spatial prior is
not feasible. For DWI, the distribution of b-values turns out to be a central
factor affecting the time-quality ratio of the DWI acquisition. An optimal
b-value distribution was determined. This helps to shorten the imaging time
without hampering the diagnostic accuracy.

Last updated on 2019-29-01 at 10:23

Share link