Date of Award


Document Type


Degree Name

Master of Science (MS)


Computer Engineering and Sciences

First Advisor

Debasis Mitra

Second Advisor

Marius Silaghi

Third Advisor

Samuel Kozaitis

Fourth Advisor

Ronaldo Menezes


Non-negative matrix factorization (NMF) is a useful method for those non-negative multivariate data. In nuclear imaging, NMF, which is also called factor analysis, is used to analyze the 4D image of positron emission tomography and single positron emission computed tomography. Normally, we use factor analysis of dynamic structure (FADS) for the dynamic reconstruction [1]. However, this reconstruction algorithm is extremely dependent on the initial data, which means the result may be unstable when the initial data is not good enough. There are several ways for the initializing, such as splined initialized (SIFADS) [2] and clustering initialized (CIFA) [3][4]. In this thesis, we have attempted different ways to improve uninitialized FADS. The dataset we used is the real human data.