An artificial intelligence approach to computerized electroencephalogram analysis
V. Jagannathan · 1981
This thesis describes the implementation and use of a rule-based EEG analysis system. The major goal of the research was to create a computer system which attempted to read an EEG in much the same way as a human electroencephalographer. Input to the artificial intelligence system (AIS) was from a technical description system (TDS) which used a combination of autoregressive and heuristic techniques to produce a succinct description of the technical attributes of an EEG. The AIS then used this semantic information to reach an overall evaluation of the EEG. The AIS was coded in the LISP programming language and used a knowledge base of EEG-related facts. The knowledge base consisted of rules of the form:^ If: certain facts are known to be true, Then: arrive at an appropriate conclusion.^ The rules could be both categorical and inexact in nature. Inexactness or fuzziness in rules was accomplished by using clause weights, by implementing the concept of a frame and by incorporating the notion of linguistic variables. The system can deduce conclusions based on these inexact rules and can also explain its' conclusions. Two distinct sets of rules were formulated. One set was designed to detect normality and abnormality in EEG data recorded from renal patients. The other set was used to conclude about sleep stages. System performance was compared with other analysis systems previously developed. A comprehensive discussion of the analysis results, the advantages in the use of AI methodology in EEG analysis, and problems which still need to be resolved are also presented.