PILOT STUDY
CSE's Asst. Prof. Xubo Song recently initiated a pilot study for the analysis and characterization of prostate Quantitative Computer Tomography (QCT) images. Researchers will develop computer-based image processing techniques to extract the prostate region from QCT scans (“segmentation”) and characterize the prostate digitally. The data from these studies will constitute preliminary data for NIH grant proposals to study important questions related to prostate health as well as biomedical imaging.
This study is significant because QCT analysis may permit not only the evaluation of QCT as a clinical screening and diagnostic tool, but should also open opportunities for the investigation of a variety of questions about prostate pathobiology. The novel image processing and analysis techniques developed in this proposal will also lead to methodologies applicable to general biomedical imaging problems.
OVERVIEW OF RESEARCH DESIGN
This project will capitalize upon the availability of data taken as part of an earlier multi-institutional NIH MrOS study centered at OHSU (Eric Orwoll, M.D., Principal Investigator). The MrOS study (the Osteoporotic Fractures in Men study) observed over 6000 men age 65 and older, from geographically and ethnically diverse populations, for about 4.5 years to examine risk factors for fractures in older men and to examine to what extent osteoporosis is associated with prostate disease. These data provide a unique opportunity for investigating the use of QCT for prostate diseases.
To develop computer-based image processing techniques that extract the prostate region from CT scans, the pilot study will use a set of selected scans to build a prostate model. The prostate regions of these CT scans will be manually delineated (“segmented”) by a radiation oncologist to form the “training set” of data from which the computerized prostate model will be developed. Then, the prostate model will be used to segment other new scans. The validity of the computer segmentation will be evaluated by another expert reader to confirm the accuracy of the model. After a prostate CT image scan is segmented, researchers will derive measurements to characterize the prostate.
 |
|
|
Computer-based Segmentation of Prostate QCT images Since (a) shape and size of the prostate vary across patients and (b) the prostate often appears as a low-contrast mass, whose interface with the rectum and bladder is not always clear, an experienced radiologist has been necessary to resolve these issues. A successful computer-based segmentation technique should reflect such ability.
|
|
ABOUT SEGMENTATION OF PROSTATE IMAGES
Segmentation (delineation) of the prostate is crucial in using prostate CT images to diagnose and research prostate diseases. To investigate the usefulness of CT images for questions such as prostate cancer screening and prostate fatty infiltration, accurate segmentation of the surface contour is the first step for subsequent quantitative characterization and visualization of the prostate.
Currently, segmentation of QCT scans is done using manual drawing software. With this software, the radiation oncologist outlines the prostate region on a stack of cross-sectional x-ray slices that constitute the prostate. This procedure is tedious and time-consuming. (An experienced radiation oncologist would need to work full-time for about 8 months to segment the 4000 CT images in the MrOS cohort, assuming 15 minutes per scan, unvalidated.) This is a major drawback to using CT images in efficiently diagnosing and researching prostate diseases. A computer-based segmentation algorithm of the prostate is needed.
COLLABORATIVE RESEARCH DRIVES EDUCATION
This research study investigates image processing issues and will likely be used in Dr. Song's graduate-level classes to illustrate the kinds of problems encountered in this field. This is basic information for SAS/EE students, who are required to gain knowledge of both signals and image processing. Other fields that touch this research include: machine learning, pattern recognition issues, image retrieval, databases, multimedia and network streaming.
__________________________________________________________________
N.B., Song's colleague, Assoc. Prof. Melanie Mitchell., is using the same data under a three-year grant from Intel to study evolutionary computing for biomedical image analysis. more...
