Knowledge-based factor analysis of dynamic nuclear medicine images.

摘要

The aim of this dissertation was to improve the processing and analysis of dynamic nuclear medicine image sequences. It addressed the following major limitations of dynamic nuclear medicine imaging: the presence of noise due to nuclear counting statistics, limited spatial resolution due to finite detector size and filtering of data, and difficulties in the definition of kinetic models and the estimation of the related parameters. Principal component analysis (PCA) was used to both reduce noise and provide an initial description of the data. In the first case, the merits of applying the PCA in images space, projection space, and volumetrically were explored. The second of use of PCA was to provide an initial solution for factor analysis (FA). Knowledge-based factor analysis (KBFA) was developed to utilize additional a priori information to transform this initial solution into a more physically realistic and physiologically meaningful solution. The proposed methods have been applied to simulated data, clinical single photon studies, and clinical research studies in positron emission tomography (PET). PCA and FA demonstrated the ability to characterize and distinguish various normal tissues as well as detect motion artifacts. Additional benefits in signal identification, noise reduction, and spatial resolution recovery arose from the application of PCA volumetrically and in projection space. Factor analysis of cerebral glucose metabolism and neuroreceptor PET studies demonstrated the ability to identify and characterize diseased tissue from normal function and specific from non-specific binding, respectively.