Matlab / Sun grid processing PDF Print E-mail
Written by Administrator   
Thursday, 19 August 2010 15:04

At the Olin Center, we originally had a half dozen Solaris computers that we used for fMRI processing. Storage and processing were done on the same box, and data moved between servers only if we ran out of space on one. A couple years ago we got a linux cluster with 28 nodes, 8 CPUs each, and a separate 100TB storage array. The whole cluster is controlled by Sun Grid Engine and I adapted our fMRI processing to fit that.

Since we've run more than 7000 studies, with more than 30,000 functional data series, we have a lot of data to process. Automatic batch processing was the best solution. Simply, I built a system to treat the processing of each subject as a cluster job, and created a manager to submit jobs to the cluster.

This allowed us to preprocess (realignment using INRIalign) and runs stats on a batch of 1500 subjects in one week.

Surgical navigation PDF Print E-mail
Written by Administrator   
Saturday, 07 August 2010 18:35

I got involved in a very interesting project at work with fMRI and presurgical mapping. A couple years ago the neurosurgery department at Hartford Hospital contacted the Olin center to form a collaboration with us. The goal was to have patients who were about to undergo brain surgery to do some simple fMRI tasks to map out language, motor, visual areas before surgery. The research has two parts: 1) help the surgeon avoid removing important areas of the brain if possible 2) get information back from the surgery to see if our estimations of language, motor, visual areas were accurate. During the surgery, the surgeon stimulates areas of the brain to 'turn off' activity in those areas. If the person can no longer speak when an area is stimulated, then that area is responsible for language.

My involvement was in the second part, comparing the computed areas of activation from fMRI with the real stimulation results. To do this, we needed to get the coordinate of the point of stimulation that actually did something. So for example the surgeon identifies an area responsible for language, and we note the location of the probe on the navigation system. Originally, the neurosurgery department was using BrainLab navigation system (which they still are), which did not directly give us coordinates back after the surgery. It was my job to attempt to decrypt the output from BrainLab to get the coordinates we wanted, however the department chose to trial the Medtronic system recently. This system gave us the exact coordinates we wanted, in image space! We will see what happens as the Medtronic trial moves on.

Observing surgery was pretty interesting. Those surgeons are amazing: no sitting, eating, or bathroom for 7 hours.

Last Updated on Thursday, 19 August 2010 15:04
gbookcards.com PDF Print E-mail
Written by Administrator   
Sunday, 31 January 2010 21:20

A long time ago, around 2002, I was working on a C++ project and I was looking for a C++ quick reference. Nothing in depth, but a single sheet of paper that contained all the common c++ syntax. Since I couldn't find anything, I decided to create a C/C++ reference card. Over the years I ended up creating a Chemistry 1, Chemistry 2, and with the help of some other people, PHP and Organic Chemistry cards. I got some ISBNs from lulu.com, printed the cards myself, and sold them on my site (http://gbookcards.com), eBay, and amazon. Unfortunately, the demand was so low for the cards, it wasn't worth it for me to pay Amazon for the right to sell them on their site. So I just let the venture fade away. Perhaps someday I will make the cards open-source and allow people to print the PDFs on demand.

Entering last semester of Masters PDF Print E-mail
Written by Administrator   
Saturday, 07 August 2010 18:19

I'm now entering my last semester at UConn in the biomedical engineering master's program with only one class left to take... and I need to finish the thesis too. Since January, I used my little 3D measuring program to measure 95 aortic valves and found some interesting results, such as: there is a significant difference in valve sizes between genders, but not between age groups. To make the paper more interesting, we also added measures obtained from different methods such as 3D surfaces and 2D automatic measuring.

My original plan when I started doing my thesis was to work on 3D active appearance models. This is slowly turning out to be more work that it will be worth. The amount of time necessary to pick the landmarks for a single 3D model is far more than for a 2D model. If a good 2D model has 50 points, a good 3D model of the aortic valve would have 20 slices, each with 50 points, for a total of 1000 points. That number can be reduced, but how many landmarks can you remove before the model is no longer usable by the AAM algorithm. It just seemed improbable, even if I could extend the 2D methods to 3D, to make this 3D method work. There was just way too much manual work necessary to create the point models before even applying them to targets.

The next thing I wanted to try was a generic approach, which seems much simpler and more computationally efficient. SPM (statistical parametric mapping) has a high dimensional warping toolbox which deforms a target to fit a template, in 3D, and it allows control of the amount of deformation. So far, its looking promising, but only testing will determine its usefulness. This will probably become the bulk of my thesis.

After the thesis and class are finished, its graduation time! Then who knows what will happen :)

Last Updated on Tuesday, 24 August 2010 00:26
Valve models from 64-slice cardiac CT scans (using active appearance models) PDF Print E-mail
Written by Administrator   
Tuesday, 29 December 2009 19:46

This project is the basis for my thesis for my masters degree at UConn. My main goal with the project was initially obtaining the data, which was more challenging than anticipated. Hospital PACS systems are generally designed to store lots of imaging data and make it available for 30 days or so. Then the images go into an archive and can be retrieved when necessary... which generally means one patient at a time. PACS is not designed to retrieve large batches of data, like the 900+ patients we got IRB approval to analyze. The 125 or so patients that I did get will be very useful for future imaging studies though.

I was interested in active appearance models (AAMs) as a method to extract 3D valve models from the CT data because of a project I did in my first semester in a pattern recognition class. I actually used active shape models (ASMs), which are very similar to AAMs, to analyze 2D brain images. It worked fairly well and was able to extract the features from most target images. It wasn't terribly robust because it didn't take into account the texture information like AAMs do. I was hoping to extend the 2D ASMs into 3D AAMs for my thesis.

Transforming the 2D statistical shape model (SSMs) algorithm, of which AAMs and ASMs are subset, to 3D is not complicated. A decision must be made to use polar coordinates, quaternions, or another method of rotation in 3D space, but the general equations and algorithm are very similar. The difficult part is in creating the training data. Creating landmarks on 2D training data might involve picking 25, 50, or more points for each training image. The position of the points is generally the same in each image, so loading a 'master' set of points onto each image and moving the points individually to each landmark is relatively easy.

When trying the 2D approach in 3D it gets a lot more complicated. A simple solution, which has been done in the past is to reslice the volume and apply 2D AAMs to each slice. The drawback to this method is that the 3D objects are of different heights and can occupy different numbers of slices. This can be overcome by normalizing the heights, however the internal structures will appear on different slices. So that method is out.

The method I initially chose was to reslice the CT volumes along the long axis of the aortic valve (the valve were interested in), and establish a set of landmark points that describe the important details of the valve. The points would be defined individually for each of the training datasets and manually moved in a 3D viewer to correct locations. The problem with this is the size of the landmark set. Two hundred points might be necessary to accurately describe the valve shape in 3D, however this is an enormous number of points and would extremely time consuming to click them all by hand.

So now I search for better landmarking methods, and better methods are not limited to using AAMs. I may continue with AAMs, but if I find a simpler feature extraction method, I will probably experiment with it.


Page 2 of 4
Copyright © 2014 gbook.org. All Rights Reserved.
Joomla! is Free Software released under the GNU/GPL License.

Who's Online

We have 1 guest online


Featured Links: