The objective of this proposal is to test two hypotheses of the imaging characteristics of positron emission tomography (PET): (1) that substantial improvements in reconstruction quantification can be obtained for positron emission tomography (PET) systems by utilizing an automated, expert system that determines the best algorithm and reconstruction parameters for quantification for a particular lesion in a particular patient; and (2) that determination of the local PSF will allow more accurate and flexible use of the system. Reconstruction is an essential component of PET imaging. Many algorithms have been developed, but the algorithm that gives the most reliable SUV measurement depends in a complicated way on many circumstances of the acquisition, including ? but not limited to ? count level, lesion size, lesion shape, lesion location, background level and structure (e.g., a lesion near the bladder versus the liver), and patient size. In addition, the quantitative response of that reconstruction depends on parameters of that approach, such as iteration number, point- spread function (PSF) model parameters, filtering, and the particular lesion. We will synthetically embed lesions of known size, shape, location, and activity concentration into an existing data set. This will allow us to know the truth and extract the response of the reconstruction. We can then compensate for this response. In addition, we can process different algorithms and reconstruction parameters to determine the best combination for each lesion in each patient. Our second approach synthetically embeds point-source data very near the lesion, as opposed to embedding a lesion of similar size. This will give us two data sets: with and without the point source. We will then reconstruct both sets and take the difference to estimate the local PSF in reconstruction space. This local PSF can be convolved in reconstruction space with the estimated lesion shape to calculate the estimated bias and noise for an ROI. This second method has the advantage that corrections and variance can be determined for arbitrarily shaped ROIs, but the disadvantage that more processing ? and perhaps error propagation ? is needed.
The specific aims of this proposal include: (i) developing and integrating the initial expert-system tools that will allow for graphical user input and for the execution of ensembles of lesions with the use of different reconstruction algorithms and appropriate ranges for reconstruction parameters; (ii) developing a new method for using embedded point sources to estimate the PSF in each patient's reconstruction as a function of reconstruction algorithm and its associated parameters as an alternative way to estimate bias and variance in SUV measurements; (iii) testing the system with phantoms that have lesions of different size, shape, and SUV values; and (iv) testing the system by embedding clinically relevant lesions of known size, shape, and location, as recommended by our physicians, into archival patient data.
This project develops a software framework for efficiently processing PET patient studies to determine the reconstruction algorithm and its parameters that best quantify each lesion in each patient. In addition, quantitative corrections for the reconstruction and the uncertainty are estimated. This tool could allow for personalized PET reconstructions and more accurate quantification of tracer studies for diagnosing and treating diseases, particularly cancers.
