Author : Alexander Mihlin
Publisher :
ISBN 13 :
Total Pages : pages
Book Rating : 4.:/5 (15 download)
Book Synopsis Statistical and Computational Techniques for GPU-accelerated PET Image Reconstruction by : Alexander Mihlin
Download or read book Statistical and Computational Techniques for GPU-accelerated PET Image Reconstruction written by Alexander Mihlin and published by . This book was released on 2018 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: Positron emission tomography (PET) is an imaging modality that can detect a contrast agent that preferentially accumulates on or inside diseased cells with concentrations as low as pico-mol/L. Since diseases typically begin on molecular and cellular levels, PET's sensitivity to fine molecular changes makes it essential for detection, staging, and treatment of oncological, cardiovascular, and neurological diseases. Moreover, PET is indispensable for basic research of biological processes, and pharmaceutical development. This dissertation presents mathematical and algorithmic techniques for increasing the safety, ac- curacy, and affordability of PET imaging. Particularly, it presents the first ever maximum likelihood expectation maximization (MLEM) algorithm for photon attenuation correction from PET emission data alone. This is the only existing technique that guarantees monotonic increase of PET image likelihood with estimation iterations. Moreover, the dissertation presents advances in stochastic modeling, inverse problems with incomplete data, numerical optimization, parallel computing, and graphics processing unit (GPU)-based formulation of the method, that reduce image estimation du- ration from 5 days to under an hour, by accelerating the algorithm by over 200-fold compared with single CPU-based formulation, and reducing its memory usage by 5-fold. Furthermore, the disser- tation shows how these advances could benefit other algorithms that model the imaging system in PET, SPECT, and CT. Particularly, it shows how they can accelerate single scatter simulation (SSS) by over 100-fold compared with single CPU-based formulation, and increase PET's geometrical sys- tem matrix compression used in the image reconstruction process by over 800-fold compared with today's state of the art methodology. Finally, using the advances described above, the dissertation presents the first ever MLEM algorithm for joint correction of photon attenuation and tissue-scatter from PET emission data alone.