| |
Gene Gindi, Ph.D.
Associate Professor of Radiology and Electrical
& Computer Engineering
Funding through the National Institute of Neurological Disorders and Stroke.
The focus of my lab, the Medical Image Processing Laboratory, is on the
reconstruction and analysis of medical images, with particular emphasis
to photon-limited modalities such as PET, SPECT and CT. The work is driven
by two questions; (1) The collected data is degraded by noise and systematic
error, and is in itself only indirectly related to the actual underlying
image. How can we devise and implement mathematical methods to solve the
inverse problem of estimating the image from the collected data? (2) The
resulting reconstructed image is then used to support tasks such as quantifying
local regions, or detecting the presence of of a weak signal buried in
the image. How can we quantify the ability of the reconstructed image
to support such tasks?
Figure 1. A PET reconstruction of
a human brain. This 2D slice shows metabolic activity in the
form of an image of a tracer distribution. The FDG tracer (F-18
deoxyglucose) maps metabolic activity. The image is reconstructed
by computer from photon counts received at hundreds of detectors. |
|
Our current focus is in SPECT (Single-Photon Emission Computed Tomography)
and PET (Positron Emission Tomography). In both, the data comprise photon
counts received at thousands of detectors surrounding the patient. The
challenge is to take this data and reconstruct it by minimizing a cost
functional that yields an image estimate optimal in some probabilistic
sense. Doing this also involves understanding the detailed physics of
the image formation model, i.e. the "forward model" relating the underlying
image to its appearance as data. Since we can characterize in detail the
source of uncertainty in the data, another challenge is to use analytical
methods to propagate this uncertainty into the reconstructed image, thus
yielding limits on task performance as a function of noise level and acquisition
time. A third goal, related to uncertainty propagation, is to use methods,
such as texture synthesis, to simulate "true" underlying images, such
that a human observer would have a difficult time deciding whether they
are viewing a real or a synthetic medical image.
While the focus of photon imaging with X-rays and gamma rays has been
in clinical medicine, these methods are beginning to be used in gene-expression
studies with small animals. The image science challenges in these areas
of biological application are quite similar to those in the clinic. Indeed,
some of our past work has involved the use of metabolic tracers in small
animal imaging.
Student Background: An ideal undergraduate should have a good engineering
math background, especially an introductory familiarity with probability
and statistics, optimization theory, and linear algebra. Useful also are
programming skills in C/C++ and MATLAB. Finally, a physics background
with an exposure to optics and introductory atomic/nuclear physics would
be a bonus if not a prerequisite. Such a student would be able to accomplish
significant work over the course of a summer. Students majoring in Electrical
Engineering, Applied Math, or Physics, as well as Computer Science Majors
with good math backgrounds, would be ideal.
Contact Information
email: gindi@clio.rad.sunysb.edu
url: http://www.mipl.rad.sunysb.edu/mipl
back to top
|
|
