Estimation of Variability in PET

Reconstructions

Dr. Ranjan Maitra

Bellcore

Wednesday, February 26,1997

3:15 pm - refreshments

3:45 pm - talk

(Please note slight deviation from usual times)

Room 327, Yost Hall

Abstract

Positron Emission Tomography (PET) is a radiologic imaging tool capable of revealing tissue biology. Reconstructing each PET image is a high-dimensional linear inverse problem. While estimating variability in these images is important, this is impractical because (unlike reconstruction) calculations can not be implemented using discrete convolution. We propose approximation strategies that use convolution- type methods, readily implemented using Fast Fourier Transforms. Results on model experiments suggest reasonable performance. The approach is extended to suggest synthetic simulation methods for estimating dispersions in functional images of biologic parameters --- a hitherto impractical proposition in PET.