A RUDN University mathematician with colleagues from Egypt, India, Poland and Saudi Arabia taught artificial intelligence to identify pathologies on an electrocardiogram. The model works with almost 100% accuracy and outperforms all previous analogues in efficiency. The results are published in Sensors.
The main tool in the diagnosis of cardiovascular diseases is the electrocardiogram. The experienced eye of a doctor can notice the smallest violations and make a diagnosis, but the longer the cardiogram, the harder it is to do so. For such purposes, to help the doctor, a computer algorithm is needed that will help not to miss the pathology. A RUDN University mathematician with colleagues from Egypt, India, Poland and Saudi Arabia built an algorithm based on machine learning for this.
“Anomalies in the creation and conduction of an electromagnetic wave due to disturbances in the functioning of the heart can be seen using an ECG study. An electrocardiogram can help identify the most common heart conditions — arrhythmias, coronary heart disease, and heart attacks. However, manual evaluation of ECG signals is time consuming and tedious. Therefore, it is important to develop an accurate methodology for automatic anomaly detection,” said Ammar Muthanna, PhD, Director of the Research Center for Wireless 5G Networks Simulation of RUDN University.
Mathematicians investigated the ECG database for the diagnosis of myocardial infarction. It includes more than 20 thousand records received from almost 19 thousand patients. RUDN mathematicians used the so-called convolutional neural networks. This is a multilayer artificial neural network in which the layers alternate with each other according to their functional purpose. Such networks use the features of the visual cortex and are aimed at pattern recognition.
The new model makes it possible to detect myocardial infarction by electrocardiogram with an accuracy of 99.2%. This is the best result among analogues. The previous record is 93.1%. In the future, RUDN University mathematicians plan to further improve the result and expand the scope of the new model.
“Our method is more reliable and more efficient than previous models for detecting ECG anomalies. In the future, we will test the proposed model on other types of signals, for example, on an electroencephalogram. In addition, we can perform optimization to select the best parameters. Finally, we can try to use more than one classifier on the same dataset and observe the results," said Ammar Muthanna, PhD, Director of the Research Center for Wireless 5G Networks Simulation of RUDN University.
RSS