Methodological advancements in data analysis and quantitative performance of positron emission tomography

Abstract

Positron emission tomography (PET) is a medical imaging modality with which neurophysiological functions can be studied. After a radio-labeled molecule is injected intravenously, it is transported by the blood stream to the target of interest. The radioactive atom of the molecule decays, resulting in two gamma rays that the PET system detects. Based on this measurement, an image can be reconstructed displaying the distribution of the radioligand, which in turn provides useful information regarding the underlying physiology.

This thesis focuses on methodological advancements in the quantification of PET data obtained from neurological studies. The work included in the thesis can be categorized into three sections.

Section 1 comprises two studies (Study I and II) and focuses on the effect of the resolution of the PET system when quantifying the data. When using PET to study brain functions in for instance the brainstem, the structures of interest are small and therefore presumably affected by resolution. The results of study I showed that, using conventional methods for PET data analysis, the difference obtained between two PET systems with different resolution can be compensated for by applying algorithms that artificially compensate for resolution-induced image artifacts. This is an interesting finding as it enables data acquired in two different PET systems to be pooled into the same analysis, without introducing significant bias. In study II, we found that using a high resolution PET system, in combination with a noise suppression technique and a semiautomatic procedure to define regions of interests (ROIs), it is possible to accurately quantify radioligand binding in very small brain structures. The procedure allows for detailed mapping of the distribution of serotonin transporter in the brainstem, and may thus be used to help elucidate the role of the serotonin system in central nervous system disorders.

Section 2 targets the definition of ROIs required for analysis of PET data. Conventionally, ROIs are defined manually by a neuranatomically trained expert on MR images acquired from the same subject. This procedure is time consuming and introduces a large amount of user interaction in the data analysis, making it prone to rater bias. In study III, it is shown that manual ROIs can reliably be replaced with automated versions provided by the software package FreeSurfer or the Automated Anatomical Labeling (AAL) template. These automated methods provide objective and reproducible analysis of PET data.

In section 3, a new method to cluster the voxels of PET images is presented. This Pair-Wise Correlation (PWC) approach groups correlating voxels, either within or across subjects. The method is used for two different purposes. First, in study IV, the PWC method isolates the signal corresponding to arterial blood in the images. This image-derived blood signal provides the possibility to quantify PET data without measuring the radioactivity level in arterial blood, which significantly reduces the invasiveness during a PET examination. Second, in study V, the PWC method identifies a disease-specific pattern of amyloid plaque in patients with Alzheimer’s disease. The pattern can in turn separate a group of AD patients from control subjects, suggesting that the PWC method may aid for early and objective detection of brain amyloid.

In conclusion, this thesis focused on validation and implementation of advanced tools for quantification of PET data. The results indicate that the methods included in this thesis provide improved quantification of PET data and can be used in clinical PET studies.