Development, validation and application of advanced neuroimaging analysis tools for in vivo neuroreceptor studies

Positron emission tomography (PET) is an imaging technology, which can be used to study neuroreceptors in the human brain in vivo. The technique estimates the regional binding of radiolabelled ligands to neuroreceptors and the data are commonly displayed as images showing the distribution of radioactivity in the brain volume. A traditional image analysis approach builds upon reduction of noise in a region of interest (ROI) by averaging radioactivity in the volume elements (voxels) of the ROI. This approach is efficient to improve the reliability of the time curves but does not allow for a detailed analysis of the entire brain. To obtain detailed three-dimensional maps of binding parameters in brain, novel approaches have been developed during recent years. The aim of the present thesis project is to examine and validate the repertoire of advanced computerised tools used to obtain parametric maps of receptor binding in basic and clinical neuroscience research. In addition, the increasing number of suitable PET radioligands, targeted for different neuroreceptor systems, calls for approaches that allow for a combined analysis of multiple receptor systems. A parametric mapping approach using wavelet filtering was evaluated in a crossvalidation design. Data from PET-studies on regional D2/D3 dopamine and 5-HT1A serotonin receptor binding in the human brain were used to compare the binding potential (BP) estimates of the wavelet-based approach and other parametric imaging approaches to the ROI-based graphical Logan analysis which was used as a reference. The approach using three-dimensional wavelet filtering was noise-tolerant and yielded BP maps with regional averages closely matching the reference values. Overall, the wavelet-based approach seemed to provide the most valid and reliable estimates across regions with a wide range of receptor densities. However, there was some loss of resolution, which may be critical for analysis of binding in small anatomical regions. Another set of parametric mapping approaches is similar to the ROI-based analyses in the sense that signal averaging is used to reduce noise. However, these approaches do not average the time-activity curves (TAC s) of spatially adjacent voxels but that of voxels having a TAC with a similar shape. A process was developed to classify voxels into a large number of groups (clusters) and thus to obtain an average TAC for voxels with a similar TAC. The classification was performed using an artificial neural network model, called the growing adaptive neural gas (GANG), which was developed as part of the thesis. Parameter estimation was performed on the average TAC s and the parameters were then back-projected to the original spatial locations of the voxels thereby providing 3D parametric maps. The approach was applied to PET images measuring D2/D3 receptor binding. The results indicate that the approach can be used to effectively reduce noise. The created parametric maps were highly detailed and the binding distribution was consistent with parametric images obtained with previous approaches. Novel technical approaches are required in combined analyses of multiple neuroreceptor systems. Such approaches have to be capable of operating on very large parametric image datasets. An initial step is the development of exploratory data-mining tools, which provide guidance as to the structure of complex multi-individual, multi-receptor datasets. For this task, an unsupervised and unbiased data-mining tool was developed and proposed. The tool includes a GANG-based clustering of multi-receptor data. The proposed approach was tested on a dataset containing BP maps of the serotonin transporter and the 5-HT1A receptors obtained in the same individuals. The outputs of the method were multi-receptor maps with potential to reveal complex relationships and tendencies in a dataset with several ligands. Such maps may have value in clinical research on multi-receptor interactions and pattern changes in the human brain. In conclusion, the present thesis has examined and extended a methodological platform that allows for additional gain of information from routinely generated data in PET studies on neuroreceptor binding. The results support application of parametric image analysis in basic and clinical research