Presented at the 1997 APIII Conference                        Return to 1997 Abstract Index

---

Content-based Image Retrieval and the Pathology Image Database
Zheng, Lei
Pittsburgh Supercomputing Center
Graduate Student Research Program
Pittsburgh, Pennsylvania

As a part of a collaborative effort between the Pittsburgh Supercomputing Center (PSC) and the University of Pittsburgh Medical Center (UPMC), our content-based image retrieval (CBIR) research project is at the center of interest, and the cutting edge of research activities. The objective is to allow searching of the huge pathology image database by image content instead of the attached text descriptors. This would make the database a precious tool in diagnosing new cases by locating similar images along with associated diagnoses, with the help of a set of CBIR tools currently under development.

The Department of Pathology, UPMC, has stored over 60,000 high-quality digital images covering almost every area of research and practice in medicine, including both gross images and microscopic images, each annotated by experienced UPMC pathologists. The number of images in the archive is increasing rapidly, and will eventually reach 4 million. That represents a data volume of a few Tbytes. The project will take advantage of PSC's extensive experience in high performance computing, as well as massive storage capacities, to extract and evaluate medically significant features using image processing, information theory, and machine learning techniques. These features will then be used as indexing keys to construct the internal database accessing structure. A Web-based interface will be employed to deliver the database content and CBIR search tools across the Internet. As a test case and a valuable stand alone application, our computer automated prostate cancer grading method correlates well with the Gleason grades assigned by UPMC pathologists.