Unit:
Κατεύθυνση Πυρηνική Φυσική και Φυσική Στοιχειωδών Σωματιδίων (ΒΑΣΙΚΗ ΦΥΣΙΚΗ)Library of the School of Science
Supervisors info:
Γεώργιος Τζανάκος Καθηγητής ΕΚΠΑ, Σακελλίου Λουκάς Αναπλ. Καθηγητής ΕΚΠΑ, Στυλιάρης Ευστάθιος Αναπλ. Καθηγητής ΕΚΠΑ
Original Title:
Μεθοδολογία για την ανακατασκευή εικόνας σε τομογράφο PET βασιζόμενη στον αλγόριθμο MLEM
Summary:
The present thesis is concentrated on the study of the convergence of the MLEM
(Maximum Likelihood Expectation Maximization) iterative algorithm . Using the
digital Hoffman brain phantom and Monte Carlo techniques, tomographic
projections are created. These projections are used from the MLEM algorithm
which reconstructs each slice of the phantom comparing it with the initial
picture. It is observed that the reconstructed images that are produced are
improved as long as the number of iteration increases reaching the best
quality reconstructed image which is altered after that iteration. Recent
researches have used the quantity NRMSD (Normalised Root Mean Square Deviation)
as quality index of the reconstructed image because it compares the data of the
two images, initial and reconstructed. According to this quantity new
stopping rules have been developed for the algorithm, focusing on sparing time
in comparison with previous methods. The present thesis presents new stopping
rules using the relative derivative of NRMSD which had a threshold. The
algorithm stops after that threshold. The right choice of threshold ensures the
quality of the best reconstructed image The results of this study concerning
the algorithm’s acceleration are presented at this thesis.
Keywords:
Positron Emission Tomography, MLEM Algorithm, Optimum Iteration, PET & MLEM, Normalised Root Mean Square Deviation (NRMSD)
