Indexing by Content
Click here for higher magnification (200K)
The world's literature has become digital. Presently dozens of journals
and textbooks are available in digital form through the Internet or on CDs.
Searching this body of knowledge for text is simple. Searching for images
is not. An image can only be found if it has a label.
Dr. David J. Foran, Assistant Professor of Pathology and Director of
the Center for Biomedical Imaging at Robert Wood Johnson Medical School,
and colleagues are developing methods whereby image databases can be queried
by content. This is being done presently by selecting a region of a digital
image on the computer screen by tradition drag-click methods and then asking
the computer to find all examples in the database which 'resemble' the selected
region. A demonstration of this process as applied to lymphoproliferative
disorders is presented below in abstract form and will be formally presented
in November, 1998, at the 'Imaging in Anatomic Pathology' meeting at the
University of Pittsburgh.
Image Guided Decision Support for Detecting
and
Differentiating Lymphoproliferative Disorders from
Peripheral Blood Smears
Lauri A. Goodell, Dorin Comaniciu, Peter
Meer,
David J. Foran
Department of Pathology, Robert Wood Johnson
Medical School and CAIP, Rutgers Univeristy
The subtle visible differences exhibited by some malignant lymphomas
and chronic lymphocytic leukemia give rise to a significant number of false
negatives during routine screening by medical technologists. Mantle cell
lymphoma is a recently described entity which is often misdiagnosed as lymphocytic
leukemia (CLL)/small lymphocytic lymphoma(SLL), or small cleaved cell lymphoma
(SCCL). A prototypical image guided decision support (IGDS) system has been
developed with the goal of aiding the medical technologist, pathology resident
or pathologist in the detection and differentiation of abnormal lymphoid
cell populations.
While traditional database systems utilize text-based information to
search through databases, the image-guided decision support (IGDS) approach
systematically searches through databases of consensus-graded medical cases
based upon the visual content of constituent pathology image records. During
preliminary, feasibility studies, the image-guided decision support prototype
automatically delineated salient biological structures from digitized microscopic
specimens and characterized the underlying pathology using a set of non-traditional
spatial and spectral signatures. A weighted mixture of these measurements
served as search criteria in a series of systematic queries which retrieved
diagnoses, correlated clinical data, and image records of consensus-graded
cases which exhibited spatial and spectral profiles which were consistent
with those of the undiagnosed case. Utilizing a unique mulitmodal fusion
agent, the system is voice activated and provides audio and graphical feedback.
The prototype provided the correct classification, based on majority logic
among retrieved cases, in approximately 86% of the trials. Studies comparing
these classification rates with those of medical technologists and pathology
residents are ongoing to determine the systems value as an educational tool.