COMPUTER-ASSISTED COMPATIBLE MEDICAL DIAGNOSIS AND CONSULTING

RAMASAMY UTHURUSAMY, Purdue University

Abstract

In this thesis, evaluation of the development and implementation of a microcomputer-based system for compatible medical consulting and diagnosis is described. The system incorporates balanced approaches to introduce and integrate a priori medical knowledge with appropriate statistical decision making theory. The organized approach is based on subsystems with classes and class-feature relationships. Through this system structure a medical data base is organized in a consistent, useful format for memory stimulus in consulting, education, and assisted decision making. The system has the capability to provide instantaneous recall of medical knowledge in the form of text, data, and relevant color pictures with sound interactively for consulting, diagnosis and education using a microcomputer controlled intelligent videodisc player. The system emphasizes the fact that the physician is the manager and through interaction it assists him in his decision making. The implemented system is easily updatable and subject to peer review. It is inexpensive, easy to use and is available as a self contained unit to each practicing physician on an individual basis. This report summarizes the experience with the system and its applications to three well structured recurring medical problems and to the microcomputer managed videodisc player in interactive medical education, consultation and diagnosis. First an overall description of the system model and its implementation is given. This is followed by brief discussions of basic consulting, automatic consulting, consulting with decision nodes, networking, intermediate classes, and decision making, using statistical pattern recognition. Next, the applications are described. The first application is the important problem of differential diagnosis of chest pain. This is a single subsystem problem and is an example application of basic consulting and automatic decision making using statistical pattern recognition techniques. Experimental results indicate that the model using the k-nearest neighbor decision rule has a better performance than the model using the multivariate Gaussian decision rule. A pragmatic approach undertaken to improve the decision making performance that effectively incorporated a priori medical knowledge is also described. This approach also handled the mixed feature problem, missing and insignificant features, using equivalence regions. It also provided for sequential diagnosis and a better presentation of a posteriori probabilities to the user. This approach established criteria for selection of subsystems and their evaluation and a procedure to estimate the a posteriori class probabilities. The second application is the differential diagnosis of bleeding disorders and anemia. This is a multiple subsystem problem and is an example application of basic consulting, automatic decision making with logical rules, networking, and sequential diagnosis. The third application is the differential diagnosis of jaundice. This is a single subsystem problem and is an example application of basic consulting, automatic decision making using logical rules and networking. The fourth application is the microcomputer controlled videodisc player to provide color pictures with sound for better consulting and diagnosis. This application enhanced the efficacy of the overall system. These applications illustrate that the implemented system, with the revolutionary and exciting possibility of color pictures along with sound and text, offers a promising new means of delivering interactive medical education, consultation and diagnosis in recorded form to individual physicians to be used at the time, place, and pace, of their convenience at low cost for better medical care and management.

Degree

Ph.D.

Subject Area

Electrical engineering

COinS