The overall objective of this project is to improve the quality of images obtained by positron emission tomography (PET) for human studies in clinical nuclear medicine and in biomedical research. This will be done by developing a fast computer method for generating images from basic photon-count data acquired by PET scanners. Two versions of the method will be developed - the first for conventional PET and the second for time-of-flight (TOF) PET. Data from TOF-PET systems contain additional information that permits better spatial localization of coincidence events, compared to conventional (non-TOF) scanners. In some clinical and research applications of PET, an individual study produces a large number of data sets in a time sequence, with each data set containing a relatively small number of counts collected over a short duration of time, and where each data set requires a separate run of an image reconstruction method to generate the corresponding volume image. This situation arises when (a) multiple images are used to determine the uptake of the radiotracer at each voxel as a function of time (time-activity curves), or (b) multiple images are used to do retrospective image-based gating of data acquired continuously in the presence of quasi-cyclic respiratory motion. Generation of a large number of volume images requires a large amount of time when conventional reconstruction methods are used. The proposed image reconstruction method is different from conventional methods in the following respects. (a) The proposed method performs non-iterative, linear operations on the data, whereas conventional methods are iterative, and non-linear. (b) The proposed method shifts the iterative phase of the computation from conventional operations on the data into a preprocessing step that is done once for each scanner-reconstruction geometry, i.e., no iterative computation for each data set. The proposed preprocessing step performs iterative computation to build a set of weight factors to be used for non- iterative operations on multiple data sets, whereas conventional methods perform iterative computation directly on each data set. (c) The proposed method does not aim to be an optimal statistical estimator for processing Poisson-distributed PET data. Instead, the method achieves the desirable property of linear processing of the data, which leads to quantitative unbiased estimates in low-count regions of the image, as required for tracer kinetic studies and for image-based gating using short-duration blocks of data. The proposed method, in common with standard statistical iterative methods, can do event-by-event processing of PET data acquired in list mode, and it can incorporate a system model that is spatially variant. The project is based on this general framework, and involves the development, implementation, and testing using simulated data, of two specific methods, one for conventional PET data and the other for TOF- PET data.
Positron emission tomography is now well established as a valuable imaging tool for biomedical research, for the diagnosis of cancer and other diseases, and for the planning and monitoring of treatment. The proposed work involves a new method for computer processing of data that is designed to improve the accuracy of PET images, especially in clinical and research applications requiring generation of a large number of images in a time sequence. The proposed work is relevant to public health, since an improvement in the accuracy of PET images would lead to more accurate diagnosis of cancer and other diseases, more accurate planning of treatment, and more accurate monitoring of the response to therapy, leading in turn to better patient outcomes.
