The goal of this proposal is to develop knowledge-based methods for interactively extracting (or segmenting), anatomic objects from medical images. The National Library of Medicine has recently released a Request for Proposals, the """"""""Visible Human Project'"""""""", which calls for the creation of a digital image library of volumetric data representing a complete normal adult human male and female cadaver. Such an image library will find wide application in all areas of medicine. However, as the NLM long-range plan recognizes, the raw image data itself will be of limited use until the relevant anatomic structures have been segmented and classified. This """"""""segmentation problem"""""""" is a major bottleneck to automated image analysis in all areas of medicine, and is becoming critical as the rapidly increasing number of medical images threatens to overwhelm human interpreters. The two fundamental hypotheses behind my approach to this problem are 1)spatial knowledge of the shape and range of variation of anatomic objects is essential for segmentation, and 2) segmentation is a difficult problem that is not likely to be solved in the near future, so any useful system must be interactive. In previous work I have developed a representation, called geometric constraint networks, that is able to capture not only the shape, but also the range of variation of many biological objects ranging from proteins to organs. I have implemented this representation in an interactive program for segmentation of organs on 2-D medical images, and a preliminary evaluation has shown that shape knowledge can speed up segmentation time by as much as a factor of ten. In the current proposal I plan to continue this work.
The specific aims are 1) to develop improved methods for representing spatial knowledge of anatomy based on geometric constraint networks 2) to develop a series of interactive prototypes for segmentation of 2-D and 3-D image datasets, and 3) to evaluate the efficacy of these prototypes in two practical applications, reconstruction of anatomy from serial cadaver sections, and radiation treatment planning. Fulfillment of these specific aims will result not only in useful tools that reduce the current image segmentation bottleneck, but also in representations that form the basis for a spatial knowledge base of anatomy. Because of the fundamental roles that anatomy and structural biology play within the medical sciences, such a knowledge base will have wide applicability in clinical medicine, teaching and basic research.

Agency
National Institute of Health (NIH)
Institute
National Cancer Institute (NCI)
Type
First Independent Research Support & Transition (FIRST) Awards (R29)
Project #
1R29CA059070-01
Application #
3460707
Study Section
Biomedical Library and Informatics Review Committee (BLR)
Project Start
1992-07-01
Project End
1997-06-30
Budget Start
1992-07-01
Budget End
1993-06-30
Support Year
1
Fiscal Year
1992
Total Cost
Indirect Cost
Name
University of Washington
Department
Type
Schools of Medicine
DUNS #
135646524
City
Seattle
State
WA
Country
United States
Zip Code
98195
Hinshaw, K P; Brinkley, J F (1997) Using 3-D shape models to guide segmentation of MR brain images. Proc AMIA Annu Fall Symp :469-73
Brinkley, J F (1993) A flexible, generic model for anatomic shape: application to interactive two-dimensional medical image segmentation and matching. Comput Biomed Res 26:121-42
Brinkley, J F; Eno, K; Sundsten, J W (1993) Knowledge-based client-server approach to structural information retrieval: the Digital Anatomist Browser. Comput Methods Programs Biomed 40:131-45
Altman, R B; Brinkley, J F (1993) Probabilistic constraint satisfaction with structural models: application to organ modeling by radial contours. Proc Annu Symp Comput Appl Med Care :492-6