Project list
Project title: Semi-automated organ segmentation from CT images
Faculty lead: Ashish Raj, PhD
Programming skills needed: C++ and/or MATLAB
Math skills needed: CS5540 level
Abstract: You will be given 3D CT image of a patient's abdomen. Your job will be to write a GUI which allows a radiologist to performed labeled click inside various organs (liver, spleen, heart, kidnes, etc). Using these points as seed points, you will perform a full segmentation of the abdominal volume, so that in the end each organ is labeled. You may use a a region growing algorithm for the initial segmentation, but it is likely that this may need to be refined using an advanced level set or graph cut based algorithm. The organs may touch each other, and there may be little or no tissue contrast between different adjacent organs. Your algorithm will need to deal with this situation. This project may be done individually or by a team of two students.
Project title: Semi-automated extraction of tumors from longitudinal CT images
Faculty lead: Ashish Raj, PhD
Programming skills needed: C++ and/or MATLAB
Math skills needed: CS5540 level
Abstract: You will be given 3D CT images of patients' abdomen containing at least 1 tumor, at sevveral time points (in months). Your job will be to write a GUI which allows a radiologist to click inside a tumor at time point 0; co-register the volume at time 0 to those at times 1, ,,, n; find the same tumor on the other images; extract each tumor using a region growing algorithm; present extracted tumor volumes as a series in the GUI; finally compute and plot the volume and diameter of the tumor at each time point. This project requires a team of two students.
Project title: Automated removal of patient-identifying information from images
Project title: Semi-automated extraction of tumors from longitudinal CT images
Faculty lead: Ashish Raj, PhD
Programming skills needed: C++ and/or MATLAB
Math skills needed: CS5540 level
Abstract: You will be given 3D CT images of patients' abdomencontaining at least 1 tumor, at sevveral time points (in months). Your job will be to write a GUI which allows a radiologist to click inside a tumor at time point 0; co-register the volume at time 0 to those at times 1, ,,, n; find the same tumor on the other images; extract each tumor using a region growing algorithm; present extracted tumor volumes as a series in the GUI; finally compute and plot the volume and diameter of the tumor at each time point. This project requires a team of two students.
Project title: Automated removal of patient-identifying information from images
Faculty lead: George Shih, MD
Programming skills needed: Familiarity with basic C or Java
Math skills needed: N/A
Project title: Wiimote browsing of radiological images
Faculty lead: George Shih, MD
Programming skills needed: Familiarity with C, possibly including device drivers
Math skills needed: N/A
Project title: Automated extraction of kidney tumor features from CT images
Faculty lead: George Shih, MD and Ashish Raj, PhD
Programming skills needed: Familiarity with C/C++, MATLAB or Java
Math skills needed: senior-year level
Abstract: Image analysis software is required for a large CT data set of kidney tumor patients, with the aim of extracting tumor-related features from the images and then using them to do machine learning, particularly how do these features correlate with clinical outcomes and other non-imaging clinical measures. This project will involve writing image processing code to extract these features automatically from the images. Some of the features are: tumor volume, whether they are inside or protrude outside the kidney, the contrast uptake properties of the tumor, etc.
Project title: Visualization software to interactively display connectivity graphs overlaid on 3D brain MRI images
Faculty lead: Ashish Raj, PhD
Programming skills needed: Familiarity with C/C++, MATLAB or Java
Math skills needed: junior-year level
Abstract: Dr Raj's lab is in the process of extracting brain connectivity networks from MRI and MR DTI imaging data. Tthese brian networks describe which region of the cortex is connected to which other regions, and to what degree. This information turns out to be very important in detecting and classifying diseases, and also in looking at the structural substrates on which the brain's computational mechanism is based. We have succeeded in extracting such networks in several ways, but one of the main problems we face is how to succinctly convey the network level information to our peers. This project would involve writing the code necessary to visualize these networks. The idea is, you are given a 3D brain image volume and a text file specifying the graph (location of nodes and their connection strength). The job is to display the network visually and overlay the 3D brain volume. For this purpose a so-called "glass brain" rendering is frequently used, where the maximum intensity projection of the brain is displayed in a semi-transparent way. For some examples of glass brain rendering see:
http://www.brainvoyager.de/BV2000OnlineHelp/BrainVoyagerWebHelp/Glass_brain_view_of_statistical_maps.htm
http://www.medicalnewstoday.com/articles/66501.php
There is no publicly available software tool for the purpose of visualising network data overlaid on glass brain rendering. In fact there is no reasonable public glass brain rendering tool whatsoever. The ones that are in public use are very poor at conveying structural information.
Project title: Creation of 3D volume of interest for average Hounsfield unit densities
Faculty lead:James Min, MD and Ashish Raj, PhD
Programming skills needed: Familiarity with C/C++, MATLAB or Java
Math skills needed: junior-year level
Abstract: 3d volumes would be segmented through a seeding algorithm, and average HU density would be calculated. this would be useful, in particular, for those with atrial fibrillation, as the appendage is an irregularly shaped structure and regions of interest (2d) are unreliable.
Project title: creation of software to measure myocardial "synchrony"
Faculty lead:James Min, MD and Ashish Raj, PhD
Programming skills needed: Familiarity with C/C++, MATLAB or Java
Math skills needed: junior-year level
Abstract: this would involve placing a virtual point in the middle of the left ventricular cavity from which thousands of vectors would fan out in 360 degrees. these vectors would cease once they hit the left ventricular endocardium. and we would repeat this process over 20 phases of the cardiac cycle. in this manner, we could see which portions of the left ventricle come in faster than others - this would be very important in patients with heart failure or s/p myocardial infarction, as "dyssynchrony" is a major problem that affects both left ventricular contraction as well as diastolic left ventricular filling.
Project title: Simulation training module for radiology procedures using TAFFI (thumb and forefinger interface)
Faculty lead: George Shih, MD and Ramin Zabih, PhD
Programming skills needed: C/C++ or Java
Math skills needed: junior-year level
Abstract: Simulation for training physicians performing interventional radiology procedures will be an integral part of the learning process. TAFFI provides an easy to use and intuitive interface to simulate radiology procedures, since the commonly used hardware (ie, catheters) often require a similar grip to manipulate. We will implement a basic training module using TAFFI for a common radiology procedure (eg, central venous catheter placement).