Home
Biography
Members
Research
Publications
News
Links
Contact
Positions
 

@

Research Opportunities for Students Who Look for Some Exposure to Research 

@

bullet

Computer-aided diagnostic (CAD) scheme for early detection of colon cancer in CT colonography

Short Description:  The long-term goal of our research is to develop a computer that diagnoses diseases in medical images as an expert radiologist does.  This involves computer vision, artificial intelligence, and computer graphics techniques.  Student will work with laboratory members to implement algorithms.

Long Description:  Colon cancer is the second leading cause of cancer deaths in the U.S.  CT colonography (CTC) is a new test used for screening for colorectal carcinoma (i.e., malignancy) through the acquisition of a CT scan of the colon.  The diagnostic performance of CTC in detecting polyps (i.e., precursors of colon cancer), however, remains uncertain, with a propensity for perceptual errors by radiologists.  If a computer can detect small polyps and show them to radiologists, this will prevent radiologists from failing to detect them.  Therefore, we are developing a computer-aided diagnostic (CAD) scheme for detection of polyps in CTC to assist radiologists in their detection task.  Our CAD scheme involves 3D medical image analysis, computer vision techniques, computer graphics techniques, and our original image-based artificial neural networks, which are inspired by digital image processing theory and the human visual system.  Thus, our CAD scheme can help radiologists detect colon cancer at an early stage.

Responsibilities:  The student will be responsible for collecting CT cases and will be involved in the development of new algorithms for a CAD scheme.  A letter of recommendation would be available at the end of the research project.

Eligibility Requirements:   Familiarity with computers and/or computer language will be useful.

  

bullet

Computer-aided quantitative measurements of lesions in medical images

Short Description:  The research aims at the development of computer-aided quantitative measurements of lesions in medical images.  This involves 3D medical image analysis and pattern recognition techniques.  Student would become familiar with diagnostic imaging and computer algorithms, and work with laboratory members to run algorithms.

Long Description:  Colon cancer is the second leading cause of cancer deaths in the U.S.  CT colonography (CTC) is a new test used for screening for colorectal carcinoma (i.e., malignancy) through the acquisition of a CT scan of the colon.  Because patient triage for follow-up of CTC or optical colonoscopy is related primarily to the size of polyps (which are precursors of colon cancer), accurate measurement is important.  Currently, a radiologist measures the single longest dimension of a polyp as the polyp size.  The measurements are, thus, subjective and have large variations among radiologists, based on their experience (i.e., inter-observer variation) or on different times for the same radiologist (i.e., intra-observer variation).  Therefore, we are developing a computer-aided quantitative measurement of polyps in CTC to reduce the variations in radiologistsf measurements.  This development involves automated segmentation and a volume measurement of polyps.  Many investigators have developed various automated segmentation methods, but accurate segmentation is still difficult for complicated patterns of polyps.  We are developing a novel method for segmenting polyps which can incorporate expert radiologistsf determination of polyp boundaries and a relationship between the physical boundaries of polyps in a phantom and their appearance in CTC.  Thus, our computer-aided scheme can help reduce radiologistsf variations and improve the accuracy of their polyp measurements.

Responsibilities:  The student will be responsible for collecting CT cases and will be involved in the development of new segmentation algorithms for quantitative measurement of polyp size.  A letter of recommendation would be available at the end of the research project.

Eligibility Requirements:   Familiarity with computers and/or computer language will be useful.

 

bullet

Computerized detection and characterization of liver lesions in CT

Short Description:  The research aims at the development of an automated scheme for detection and characterization of liver lesions in CT, which involves time-series analysis and diagnostic imaging.  The student will work with laboratory members to implement and run computer algorithms.

Long Description:  Liver cancer is the most common abdominal malignancy worldwide, and its prevalence is increasing in the U.S.  Contrast-enhanced hepatic CT is used for detection of liver tumors.  A dual blood supply to the liver results in different enhancement patterns between liver tumors and normal parenchyma in three major phases of contrast enhancement:  arterial phase, portal venous phase, and equilibrium phase.  In the arterial phase, hypervascular tumors are enhanced, whereas in the portal venous phase hypovascular tumors are enhanced.  The difference between malignant and benign tumors can be seen in the equilibrium phase.  The most common malignant tumor is hepatocellular carcinoma (HCC), which often arises in patients who have cirrhosis or hepatitis B/C; benign tumors include hemangioma, adenomas, and focal nodular hyperplasia (FNH).  Based on the dynamic tumor response patterns, we are developing a computer-aided diagnostic (CAD) system for detection and characterization of liver tumors in CT to assist radiologists in their decision-making.  This development involves automated segmentation of the liver, detection of potential tumors by use of pattern-recognition techniques, and characterization of tumors by use of texture and kinetic analysis.

Responsibilities:  The student will be responsible for collecting CT cases and will be involved in the development of new algorithms for a CAD scheme.  A letter of recommendation would be available at the end of the research project.

Eligibility Requirements:   Familiarity with computers and/or computer language will be useful.

@

bullet

Development of machine learning for medical image analysis

Short Description:  The research aims at the development and understanding of machine learning for medical image analysis.  We have developed a new machine-learning technique inspired by the human visual system.  The student will work with laboratory members to implement and evaluate machine learning.

Long Description:  In many computerized schemes, aspects of the method require training with data; thus, machine learning plays an essential role in these schemes.  A major limitation of current machine-learning techniques is the necessity for a large number of training cases, e.g., 500 cases, so that an adequate performance is obtained.  This limitation becomes very serious in medical applications, because collection of a large number of abnormal cases is very difficult.  For example, a review of images from 714,000 patient examinations is required for collection of just 500 lung cancer cases.  In order to overcome this limitation, we are developing a novel machine-learning technique which is inspired by digital image-processing theory and the human visual system.  Unlike other machine-learning methods, ours is capable of learning image data directly, and it functions like human visual pattern recognition.  We are investigating and analyzing our machine learning to bridge the gap between human learning and machine learning.

Responsibilities:  The student will be responsible for analyzing data and will be involved in the development of new machine-learning algorithms.  A letter of recommendation would be available at the end of the research project.

Eligibility Requirements:   Familiarity with computers and/or computer language will be useful.

@

bullet

Investigation of a machine learning technique inspired by the human visual system that can be trained with a small number of cases

Project Description:  In many computerized schemes, aspects of the method require training with data; thus, machine learning plays an essential role in these schemes. A major limitation of current machine-learning techniques is the necessity for a large number of training cases, e.g., 500 cases, so that an adequate performance is obtained. This limitation becomes very serious in medical applications, because collection of a large number of abnormal cases is very difficult. For example, a review of images from 714,000 patient examinations is required for collection of just 500 lung cancer cases. In order to overcome this limitation, we are developing a novel machine-learning technique which is inspired by digital image-processing theory and the human visual system (Suzuki K et al. IEEE Trans Signal Proc 2002; Suzuki K et al. IEEE Trans Patt Anal & Mach Intell 2003; Suzuki K et al. Medical Physics 2003; Suzuki K et al. IEEE Trans Med Imag 2004, Suzuki K et al. Academic Radiol 2005; Suzuki K et al. Medical Physics 2006).

Specific aim:  Unlike other machine-learning methods, our novel machine-learning technique is capable of learning image data directly, and it functions like human visual pattern recognition.

Methods:  Our result of the analysis of the internal representations of our machine learning is reminiscent of the receptive fields of various simple units in the cat and monkey cerebral cortex discovered by Hubel and Wiesel (Hubel DH et al. J. Physiology (London) 1962; Blakemore C et al. Nature 1970). We are investigating and analyzing our machine learning to bridge the gap between human learning and machine learning.

@

bullet

gVirtual dual-energy radiographyh: Improved chest radiographs by means of rib suppression based on a massive-training artificial neural network

Project Description:  Lung cancer continues to rank as the leading cause of cancer deaths in the U.S. Lung nodules (i.e., potential lung cancer) in chest radiographs (CXR) can be overlooked in 12% to 90% of cases with nodules that are visible in retrospect (Austin J et al. Radiology 1992). Eighty-two to 95% of missed lung cancers are partly obscured by overlying bones such as ribs and/or a clavicle. To address this issue, dual-energy (DE) subtraction imaging has been developed for separating bones and soft tissue in CXRs. In spite of its great advantages, a limited number of hospitals use DE radiography systems, because specialized equipment is required. Also, the radiation dose can be greater than that for standard chest radiography in some cases. In addition, DE images have an increased noise level.

Specific aim:  The aim of this research project is to develop an image-processing technique for separating soft tissue from ribs in CXRs. The student will work with laboratory members to run software for virtual energy radiography.

Methods:  gVirtual Dual-Energy Radiographyh for separation of ribs from soft tissue in CXRs consists of a multi-resolution massive-training artificial neural network (MTANN; /émtæn/) (Suzuki K et al. IEEE Trans Med Imag 2005) and multi-resolution decomposition/composition techniques. An MTANN is a trainable, highly nonlinear pattern-recognition technique, consisting of a linear output artificial neural network (ANN) model (Rumelhart DE et al. Nature 1986) that is capable of operating on image data directly. Our scheme has two phases, i.e., a training phase and an application phase. In the training phase, the multi-resolution MTANN is trained with input CXRs and the corresponding gteachingh bone images obtained with a DE subtraction radiography system for learning the relationship between them. In the application phase, the trained multi-resolution MTANN provides a gbone-image-likeh image where ribs are enhanced when a CXR acquired with a standard radiography system is entered. The bone-image-like image is subtracted from the original CXR to produce a "soft-tissue-image-like" image where ribs are suppressed. Thus, virtual DE radiography provides a soft-tissue image and a bone image from a CXR acquired with a standard system. The major advantages of virtual DE radiography compared to DE radiography are: (1) no additional radiation dose to patients is required, and (2) no specialized equipment for generating DE x-rays is required. We will analyze and evaluate the virtual DE radiographs compared with ggold standardh DE radiographs.

@

NSF REU Program in Medical Informatics (MedIX) at DePaul University and University of Chicago

The NSF REU Program in Medical Informatics (MedIX) at DePaul University and University of Chicago enters its eighth year. We are looking for bright undergraduate students to get involved in research in the area of Medical Informatics for the summer of 2013. Participants will receive a stipend of $5,000 plus travel support to/from the REU site, travel support to present summer work at a conference, and subsistence allowance.

Important Dates -- January 9th, 2013: Application submission opens; March 1st, 2013: Application submission deadline; March 21st, 2013: Notification of decision; March 28th, 2013: Confirmation of participation; June 10th, 2013: MedIX Program Orientation; August 16th, 2013: MedIX Program last day.

Statistics on the previous REU MedIX Program (2005-2010): 81% students had at least one research publication; 39 publications (4 journal papers, 25 conference papers, 10 extended abstracts); 3 honor theses and senior projects; 4 fellowships to support graduate school; and 1 Computing Research Association (CRA) honor mention for outstanding undergraduate research.

We would very much appreciate you sharing this information with any students who you think would qualify and would be interested. For the application form and additional information, please visit the MedIX website at  http://facweb.cs.depaul.edu/research/vc/medix/2013/index.htm or contact Dr. Daniela Raicu at draicu@cs.depaul.edu.

 

Call for papers: A special issue of  Neurocomputing on "Advanced Computing for Image-Guided Intervention"

Call for papers: A special issue of Computerized Medical Imaging and Graphics (CMIG) on "Machine Learning in Medical Imaging (MLMI)"

Call for papers: International Workshop on Machine Learning in Medical Imaging (MLMI 2014) in conjunction with MICCAI 2014 in September in Boston, USA

A book entitled "Computational Intelligence in Biomedical Imaging" has been published by Springer 

A book entitled "Machine Learning in Computer-Aided Diagnosis: Medical Imaging Intelligence and Analysis" has been published by IGI Global 

Copyright © 2006-2013 The Kenji Suzuki Lab, The University of Chicago.