Electroencephalographic staging of hepatic encephalopathy by an artificial neural network and an expert system.

AIM OF THE STUDY: To provide an objective EEG assessment of hepatic encephalopathy (HE), we set up and tested an entirely automatic procedure based on an artificial neural network-expert system software (ANNESS). PATIENTS AND METHODS: A training set sample of 50 EEG (group A) and a test sample of 50 EEG (group B) of 100 cirrhotic patients were considered. The EEGs had been visually classified by an expert electroencephalographer, using a modified five-degree Parsons-Simith classification of HE. The efficiency of the ANNESS, trained in group A, was tested in group B. RESULTS: Both the ANNESS and the visually-based classifications were found to be correlated to liver insufficiency, as assessed by the Child-Pugh score (Spearman's coefficient rho=0.485, P<0.0001; rho=0.489, P<0.0001, respectively) and by the biochemical indexes of hepatic function (bilirubin: rho=0.31 vs. 0.27; albumin: rho=-0.13 vs. -0.18; prothrombin time rho=-0.35 vs. -0.52). The classifications were found to be correlated to each other (rho=0.84 P<0.0001, Cohen's kappa=0.55). However, the ANNESS overestimated grade 2 EEG alterations. CONCLUSION: An ANNESS-based classification of EEG in HE provided data comparable with a visually-based classification, except for mild alterations (class 2) that tended to be overestimated. Further optimization of automatic EEG staging of HE is desirable, as well as a prospective clinical evaluation.

An artificial intelligence approach to classify and analyse EEG traces.

We present a fully automatic system for the classification and analysis of adult electroencephalograms (EEGs). The system is based on an artificial neural network which classifies the single epochs of trace, and on an Expert System (ES) which studies the time and space correlation among the outputs of the neural network; compiling a final report. On the last 2000 EEGs representing different kinds of alterations according to clinical occurrences, the system was able to produce 80% good or very good final comments and 18% sufficient comments, which represent the documents delivered to the patient. In the remaining 2% the automatic comment needed some modifications prior to be presented to the patient. No clinical false-negative classifications did arise, i.e. no altered traces were classified as 'normal' by the neural network. The analysis method we describe is based on the interpretation of objective measures performed on the trace. It can improve the quality and reliability of the EEG exam and appears useful for the EEG medical reports although it cannot totally substitute the medical doctor who should now read the automatic EEG analysis in light of the patient's history and age.