Rob Toth – Segmentation methods with MRI scans


(music playing) Rob Toth: What we do is we try to do cancer
research in detecting cancer, so basically, we mainly focus on prostate and breast cancer,
but we’re expanding that, and every year, there’s about 200,000 new cases of prostate
cancer and a lot of these cases, what happens is you use simple what’s called PSA levels
in the prostate and you use that to find out if there is cancer or not. The current method
is to have trained radiologists spend a lot of their time trying to manually look at these
images and say where they believe the cancer is, so this is very time consuming and it’s
also, there’s a lot of interexpert variability because two different radiologists might have
differing opinions as to whether a certain region of the prostate actually has cancer,
so our lab is trying to use the images and quantitatively detect where we believe the
cancer is with certain probabilities, and if you think there might be cancer, you do
a biopsy, which is invasive and painful, so what our lab does is we have collaborators
at UPenn and Harvard and India and all over the place and we take MRI images of all different
modalities and we try to use those MRI images to detect the cancer so that, for example,
in surgery you can say there’s an 80% chance that cancer exists at this region of the prostate
or breast so that you don’t need to do biopsies and a lot of times what apneas is that the
biopsies are randomly selected so for example, you would take six random regions in the prostate
and just poke it in and do a biopsy and hopefully you find the cancer or don’t find the cancer,
so our labs are trying to use image analysis method to advance image analysis methods and
stuff that we’ve developed ourselves to try to detect the cancer, so that’s kind of an
overview for what our lab does and what I do is, my specific job is the first step,
which is finding the prostate, or finding the breast, or finding the region of the brain
you want, and then I give those images to the next person in the lab and they use those
regions of the image to detect the cancer. We do most of our actual programming in the
MATLab computer programming suite and it’s a very common research tool and we basically
have used state of the art methods to take a certain image and extract features from
that image, so for example, at every pixel in the image, we can calculate the mean, we
can calculate the difference between two pixels, the gradient, a lot of different mathematical
features and we use highly advanced pattern recognition tools to see which of those features
show the cancer, so for example, you might see that the variance at any given pixel is
highly indicative of cancer, so we would pick that feature. My professor is Dr. Anant Madabhushi, and
he’s actually one of the youngest professors in the Engineering School. He received his
PhD from UPenn in I believe ’04 or ’05, so he’s probably been here as long as I have,
and our lab only started a few years ago with just two or people, but now its expanded to
about 15 to 20 people. He likes to push his students very hard in a good way and for example,
I approached him at the end of my sophomore year, looking for research opportunities,
and I assumed I would just be helping other grad students or administrative work, but
he immediately started me on developing my own algorithms and trying to implement the
current state of the art algorithms and I guess the goal is to develop the software
that is needed for the next generation of radiologists. Undergrad research is an invaluable
experience and everybody should do, and it pretty much defined my academic career here
at Rutgers. (music playing)

Leave a Reply

Your email address will not be published. Required fields are marked *