Presented at the 2000 APIII Conference Return to 2000 Abstract Index
MORPHOMETRIC ASSESSMENT OF LIVER FIBROSIS USING AN INTERNET BROWSER-BASED JAVA APPLET
University
of Pittsburgh
Pittsburgh, Pennsylvania
Michael Nalesnik,
MD
Xiao-Ming Zeng, Anthony Demetris, Michael Nalesnik
Department of Pathology, Division of Transplant Pathology,
University of Pittsburgh, Pittsburgh, PA
Diagnostic histopathology is in transition from a purely expert-based system into an image analysis-based system. At present, diagnostic Surgical Pathology reports remain based on individual pathologic expertise, but the diagnostic line is often supplemented by a semiquantitatve recapitulation of relevant findings. This trend has been fostered by the introduction of numerous scoring systems that deconstruct the diagnosis into components and assign numerical values reflecting the extent of alterations for each component. This approach provides data that may be valuable in longitudinal studies such as disease evolution, or in cross-sectional studies such as relative drug effects. Central to this approach is the standardized use and interpretation of such scoring systems.
One approach to standardizing evaluation of individual histopathologic components is by image analysis. Commercial image analysis software packages are expensive and generally have a long learning curve. A problem-oriented approach that emphasizes ease of use and that is tailored to the study of specific histopathologic alterations may surmount this barrier. Such a system should provide for local file input and output and for basic image manipulation using an uncluttered interface that clearly guides the user through a pathology-specific operation. We are developing such an approach to address the quantitation of fibrosis in liver biopsies. The variable of fibrosis has significance in grading systems for hepatitis and cirrhosis, two classes of disease that are often associated with long term followup and multiple biopsies. Fibrosis is often assessed subjectively or semiquantitatively using one of several simple histochemical stains that imparts a specific color to fibrous tissue.
We are developing an Internet compatible, platform-neutral Java applet for this purpose. This is being programmed with the Java Development Kit (JDK)1.2 in the Oracle Jdeveloper 3.1 IDE. We incorporated a Java image analysis package available free from the NIH in order to minimize the workload involved with the setup of image processing. The applet is being developed in three interrelated phases. Phase 1 centers on functional development of the applet itself for pathology specific image processing. Phase 2 will address Java security issues in order to circumvent the limitations of Java applet-based local file operations. The third phase will lead to production of a Servlet on the web server to incorporate image information from the end users into a database for future analysis.
At present, a prototype Java applet has been developed and this contains a function for the calculation of the ratio of fibrous to nonfibrous tissue in a liver biopsy specimen. The process is designed to work with a trichrome-stained specimen and defaults to evaluating the ratio of fibrous to nonfibrous tissue in the entire tissue sample (excluding non-tissue background areas) based on color analysis. Since the actual color results of this stain vary from laboratory to laboratory, provision has been made to allow the user to select the desired background or foreground color based on direct image manipulation. This process is iterative so that the user may make multiple selections until the desired area has been highlighted. The fibrosis ratio is calculated accordingly. The applet is also capable of calculating the fibrosis ratio based on a portion of the image selected by the user. The applet is a component of the Transplant Pathology Internet Services (TPIS) website and can be accessed from http://tpis.upmc.edu/tpis/ImageAnalysis/. This and other small, problem oriented applets will generate reproducible supplemental pathologic data that may ultimately provide clinically significant information.