Quantitative image reconstruction methods for low signal-to-noise ratio emission tomography

Abstract:

Novel radionuclide therapies such as radioembolization with Y-90 loaded microspheres and targeted therapies labeled with Lu-177 offer a unique promise for personalized treatment of cancer because imaging-based, pre-treatment assessment can be used to determine administered activities which deliver absorbed doses to lesions while sparing critical organs. At present, the dose-effect relationships required for personalized planning are not well established, due primarily to inaccuracies in quantitative imaging. While radionuclides for therapies have attractive characteristics for the cancer treatment, imaging such radionuclides is challenging and complex. The objective is to develop methods for accurate quantitative imaging of lesions and normal organs. We focus on quantitative emission tomography of two widely used radionuclides, Lu-177 and Y-90, that have challenges associated with low count-rates. For Y-90 PET, we propose a formulation that relaxes the conventional image-domain non-negativity constraint by instead imposing a positivity constraint on the predicted measurement mean that demonstrated improved quantification in simulation patient studies. We also propose an image reconstruction method with trained regularizer for low-count emission tomography. Specifically, we include the convolutional neural networks within the iterative reconstruction process arising from an optimization problem. We further extend the regularized reconstruction method by incorporating anatomical information into the trained regularizer.

Chairs: Professors Jeffrey A. Fessler and Yuni K. Dewaraja

Remote Access: https://bluejeans.com/472989117