ARTIFICIAL NEURAL NETWORK PREDICTS MORTALITY IN PATIENTS WITH ACUTE MYOCARDIAL INFARCTION Jarosław Drożdż1, Ireneusz Zbiciński2, Małgorzata Wagner2, Jarosław Kasprzak1, Maria Krzemińska-Pakuła1 1 Akademia Medyczna w Łodzi, Łódź; 2 Politechnika Łódzka, Łódź
Artificial Neural Networks (ANN) can be defined as a parallel processing structure similar to a model of biological nervous system especially human brain. Basic elements of the neural network, neurones, are connected to each other via weights. Performance of the neural network depends on weights strengths and their structure which are adjusted during the learning process. The user of ANN does not have to develop algorithms in the form of programs: the network learns itself on the basis of examples: sequences of the inputs and related outputs. A study aimed at application of classical static neural modelling to predict mortality of patients with acute myocardial infarction. Clinical database of 568 patients hospitalised between October and December 1999 in 23 centres in Poland with in-hospital mortality of 15% (87 patients) was used. Gender, age, risk factors for atherosclerosis, treatment (thrombolysis, aspirin, heparin, nitroglycerine) and complications appearing during hospitalisation were evaluated by 25-question form. Classical feed-forward multilayer perceptron, 13 neurones in input layer (13 variables were selected as being potentially useful in the prediction of mortality), 10 neurones in one hidden layer and 1 neurone in the output layer was used. The net was working in continuous mode. The network was trained on the basis of 480 cases while ANN validation file contained 88 cases. 250 learning epoch led to excellent agreement between predicted and training data (one error only). ANN performance for validation set showed overall accuracy of correct predictions of in-hospital mortality equal to 90%. Concusions: Artificial neural network is capable to predict in-hospital mortality in patients with acute myocardial infarction with high accuracy. We recommend application of the ANN to help making a decision about further treatment of individual patient.