CCE Theses and Dissertations

A Method for the Application of Computer Analytic Tools to Clinical Research: Neural Networks Analysis of Liver Function Tests to Assist in the differential Diagnosis of Liver Disease

Date of Award

1997

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Graduate School of Computer and Information Sciences

Advisor

S. Rollins Guild

Committee Member

Marlyn Kemper Littman

Committee Member

Michael J. Laszlo

Abstract

There is a great deal of effort by medical educators in the instruction of the use of computers for clinical practice but little training in the methods and utility of computer applications in clinical research. In this paper I present a method for the application of computer analytic tools to clinical research. The steps described lead the student through the standard clinical research process and demonstrate the decision making that must take place in each step of any clinical research program. This example is based upon a valid and unique research question relating to my particular field of medical hepatology. Liver disease represents a significant cause of morbidity and mortality throughout the world. In order to significantly reduce the impact of liver disease it is essential that such disorders be recognized with great accuracy early on in their course. The diagnosis of liver disease is frequently difficult and expensive even for specialists in this area.

Standard liver function tests have the potential of assisting the physician as accurate, reliable, inexpensive, and easily accessible tools for the differential diagnosis of liver disease. However, the current applications of liver function tests for this purpose offer only limited value to the clinician in terms of reliability and validity. Clinical research efforts aimed at improving upon the precision of liver function tests through such techniques as test panels, test ratios, multivariate statistical methods and the applications of traditional expert systems have all had limited success and acceptance. The research described in this paper has resulted in the development of a probabilistic neural network program that was able to classify 109 sets of liver function tests into one of eighteen possible diagnostic categories with a precision of over 90%. The neural network developed as a result of this research should serve as an efficient tool for the clinician in the management of patients with liver disease. It should also act as a stimulus for further research in the application of neural network tools to clinical medical research.

This document is currently not available here.

  Link to NovaCat

Share

COinS