Fraunhofer-Gesellschaft

Publica

Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Kinetic analysis of dynamic MP4A PET scans of human brain using voxel based nonlinear least squares fitting

 
: Hohmann, C.

Köln, 2009
Köln, Univ., Diss, 2009
URN: urn:nbn:de:hbz:38-28985
Englisch
Dissertation
Fraunhofer SCAI ()

Abstract
Dynamic PET (Positron Emission Tomography) involving a number of radiotracers is an established technique for in vivo estimation of biochemical parameters in human brain, such as the overall metabolic rate and certain receptor concentrations or enzyme activities. 11C labeled methyl-4-piperidyl acetate (MP4A) and -propionate (MP4P) are established radiotracers for measuring activity of acetylcholine esterase (AChE), which relates to functionality of the cholinergic system. MP4A kinetic analysis without arterial blood sampling employs a reference tissue based "irreversible tracer model". Implementations can be region or voxel based, in the second case providing parametric images of k3 which is an indicator of AChE activity. This work introduces an implementation of voxel based kinetic analysis using weighted Nonlinear Least Squares fitting (NLS), which is fast enough for standard PCs. The entire workflow leading from reconstructed PET scans to parametric images of k3, including normalization and correction for patient movement, has been automatized. Image preprocessing has been redefined and fixed masks are no longer required. A focus of this work is error estimation of k3 at the voxel and regional level. A formula is derived for voxel based estimation of random error, it is based on residual weighted squared differences and has been successfully validated against simulated data. The reference curves turned out to be the main source of errors in regional mean values of k3. Major improvements were reached in this area by switching from fixed to adaptive Putamen masks and raising their volume from 5.4 to 12.5 or 16 ml. Also, a method for correcting reference curves obtained from nonideal reference tissues is presented. For the improved implementation, random error of the mean k3 of a number of cerebral regions has been assessed based on PET studies of 12 human subjects, by splitting them in two independent data sets at the sinogram level. According to this sample, absolute standard errors of 0.0012 in most cortex regions and 0.0053 in Hippocampus are induced by noise of voxel based activity curves, while errors of approximately 0.0025 and 0.0050 are induced by noise of the reference curves. Different types of systematic as well as noise-induced bias have been investigated by simulations; their combined effect on the computed k3 was found below 3 percent. The implementation is available as a modul of the VINCI software package and has been used in clinical studies on Parkinson's Disease and Alzheimer Dementia.

: http://publica.fraunhofer.de/dokumente/N-199338.html