Design of a Decision Support System to Diagnose and Predict Heart Disease using Artificial Neural Network; a case study (Ayatollah Golpayegani Hospital in Qom)

Abstract

Background: Considering the prevalence of cardiovascular diseases in Iran and the high rate of death caused by these diseases, correct prediction of patients' situation is important. So, it is necessary to use the prediction models with minimum error and maximum reliability. Artificial neural network (ANN) was used to evaluate the patients who had myocardial infarction or congestive heart failure. Methods: In this study, data of 497 patients were extracted from medical records who they were hospitalized in Ayatollah Golpayegani Hospital in Qom in 2018. In this regard, 19 important features of these profiles were extracted and a particular type of ANN called the Multi-Layer Perceptron (MLP) with Back propagation algorithm was used to evaluate the status of patients with heart failure. The sigmoid transfer function and tangent sigmoid transfer function were selected and trained using 19 neurons in the input layer, 6 neurons in the middle layer, and 75 % of the existing data. Neural network training was done using Matlab software. Results: The mean square error was 0.35 prior to data normalization, which reduced to 0.04 after data standardizing using the minimum and maximum method. The accuracy of the predictive model reached 89.50 % on the validation dataset. Considering the high sensitivity and specificity of the predictive model, it seems to have a good predictive power to classify patients accurately. Conclusion: This study, a neural network model was developed, which could predict heart failures accurately. The prediction is based on the use of a series of individual and clinical variables such as age, gender, shortness of breath, changes in blood pressure, and some blood tests. In this study, we tried to use important and low-cost factors for predicting heart disease, so that all people can be aware of their diseases with a lowest cost.

Keywords

• Heart disease
• Data mining
• Neural network model