|Ph.D Student||Eitan Hadar|
|Subject||Optimal Locally Adjustable Filtering of PET Images by a|
|Department||Department of Industrial Engineering and Management||Supervisor||Professor Emeritus Ben-Tal Aharon|
The goal of this thesis is to improve the resolution of images produced by Positron Emission Tomography (PET) scanners.
Known existing methods of 3D image reconstruction are using post-processing or inter-processing filtering, were the same filter is applied to the entire image space. Our Optimal Locally Adjustable Filtering (OLAF) approach is based on convolving the source data image with a dynamic Metz kernel function with locally adjustable degree and size. OLAF is searching for the best kernel to be applied on every particular “data cell”.
The dynamic filter kernel parameters are chosen as the global optimal solution of a specific objective function which is comprised of two terms; the goodness-of-fit (GOF) of the squared residuals and a negative entropy term, which is classically used as a measure of non-uniformity.
The efficiency of OLAF algorithm was tested on several PET data sets (simulated, phantom and clinical) and demonstrated an improvement both in contrast and uniformity, compared to usual fixed post-processing filtering methods. There were no cases when the algorithm “spoiled” the input image. Moreover, OLAF reduces the number of False Positives (FP) and clarifies other regions, which were not considered to be “malignant” as in False Negative (FN).