Artificial neural networks for assessment of patients with suspected acutemyocardial infarction

Abstract: An early diagnosis within the first hours after onset of symptoms is essential for the optimal treatment of patients admitted with chest pain of suspected cardiac origin. In the majority of patients, the 12-lead ECG is nondiagnostic on admission. In these patients, ruling-in and ruling-out of acute myocardial infarction (AMI) must be based on repeated measurements of biochemical infarct markers.In the present thesis the potential usefulness of artificial neural networks (ANN) for assessment of this type of patients was investigated:The performance of an ANN-based decision-support algorithm (ANN-algorithm) was evaluated and compared to three experienced clinicians. The ANN-algorithm detected Ami, and predicted infarct size earlier than the clinicians.A methodology was developed for selecting, training and estimating the performance of adequate ANN structures and for incorporating them with algorithms that were optimized for supporting clinical decision making related to early assessment of patients with suspected AMI.Methodological and epidemiological aspects of the transferability of ANN-algorithms, as key-components for classification in decision-support systems were investigated using Monte-Carlo simulation techniques. Methods for tuning such algorithms to fulfil medical requirements, and for enabling the use of laboratory measuring instruments with differences in analytical performance, were developed and evaluated.A study was performed to investigate whether the diagnostic performance of an ANN-algorithm could be further improved by adding parameters derived from early nondiagnostic ECG recordings.An algorithm, based on three co-operating ANNs, for the early graded prediction of ichemic myocardial damage was developed in order to provide new information of potential value in theclinical assessment of chest pain patients with nondiagnostic ECG.