Medical diagnostic systems: a case for neural networks

Abstract

Recent advances in computer technology offer to the medical profession specialized tools for gathering medical data, processing power, as well as fast storing and retrieving capabilities. Artificial intelligence (AI), an emerging field of computer science is studying the issues of human problem solving and decision making. Furthermore, rule-based systems and knowledge-based systems that are other fields of AI have been adopted by many scientists in an effort to develop intelligent medical diagnostic systems. In this study artificial neural networks (ANN) are introduced as a tool for building an intelligent diagnostic system; the system does not attempt to replace the physician from being the decision maker but to enhance ones facilities for reaching a correct decision. An integrated diagnostic system for assessing certain neuromuscular disorders is used in this study as an example for demonstrating the proposed methodology. The diagnostic system is composed of modules that independently provide numerical data to the system from the clinical examination of a patient, and from various laboratory tests that are performed. The examination procedure has been standardized by developing protocols for each specialized area, in cooperation with experts in the area. At the conclusion of the clinical examination and laboratory tests, data in the form of a numerical vector represents a medical examination snapshot of the subject. Artificial neural network (ANN) models were developed using the unsupervised self-organizing feature maps algorithm. Data from 71 subjects were collected. The ANN models were trained with the data from 41 subjects, and tested with the data from the remaining 30 subjects. Two sets of models were developed; those trained with the data from only the clinical examinations; and those trained by combining the clinical and the laboratory test data. The diagnostic yield that was obtained for the unknown cases is in the region of 73 to 93% for the models trained with only the clinical data, and in the region of 73 to 100% for those trained by combining both the clinical and laboratory data. The pictorial representation of the diagnostic models through the self organized two dimensional feature maps provide the physician with a friendly human–computer interface and a comprehensive tool that can be used for further observations, for example in monitoring disease progression of a subject.