A Generalized Methodology For Partial Volume Correction In Emission Tomography Y. Ma, O. Rousset and A.C. Evans 1 Montreal Neurological Institute and McGill University Abstract We describe a generic methodology to correct 3-D partial volume effects in clinical emission scans. It is based on prior knowledge of tracer biodistribution and tomographic imaging characteristics. We derive this information from registered and segmented MR/CT data. Two fast numerical algorithms are then used to estimate structure-specific recovery coefficients and activity spillover contributions. In this work we evaluate our method using MR/PET data acquired from a 3-D brain phantom. It is made of a human skull and plastic chambers to emulate radiotracer uptake in neuroreceptor imaging studies. Regional activity values among striatal structures are typically underestimated by 20-45 % depending on their spatial location. After correction they are restored to within 5 % of the true concentration. We also add several automatic steps to increase computational efficiency and simplify its usage in clinical environment. I. I NTRODUCTION Partial volume effects originate primarily from the limited 3-D image resolution in PET/SPECT systems. This is a big problem because of irregular and dynamic activity distribution in the body. It will produce spatially variant imaging distortions and non-stationary bias and variance in time activity curves. There exist many interests in image restoration methods to solve partial volume problems. In recent years it has been a common practice to analyze tomographic data using regional templates created from multimodality images [1]. This gives valuable information with proper image segmentation. We have seen increased use of correlated MR data in Bayesian image reconstruction [2, 3, 4]. This is the most fundamental approach which reconstructs each frame iteratively. It is not in routine clinical use due to high computational cost. 1 A summary submitted to 1998 IEEE Medical Imaging Conference. It is more desirable to employ non-iterative methods for the in-vivo correction of the 3-D partial volume effects in regional functional data. Some early algorithms rely on simple calculations in image space [5, 6]. However they require explicit estimates of background activity. While possible in some cases this is not always applicable in general clinical studies. We have developed an elegant method free from any unrealistic assumptions [7]. It works by estimating the magnitude of pure recovery and activity spillover between different functional entities in a given set of ROIs. In this paper we describe many improvements implemented to automate our method and make it independent of a particular tomograph. II. COMPUTATIONAL METHOD It is known that the observed activity within a particular tissue is the weighted sum of the true activity from all the active tissues in the biological system. Assume there are N different tissues participating in the imaging experiment each with a relative homogeneous tracer uptake. Let vectors and denote the mean and variance measured from any sets of ROI templates. One can easily derive this relationship: (1) where is the transfer matrix and the true activity of each tissue. Note that the weighting factors depend on activity distribution but independent of its concentration. This then represents a set of linear equations which can be solved to obtain the true regional values: (2) where is the inverse matrix of and the variance of the corrected data. Both of them depend on 3-D image resolution and data analysis strategy.