Using neural networks to predict cardiac arrhythmias

Cardiac arrhythmias are classified by abnormal activities in the heart. These abnormalities can be analyzed by an electrocardiogram (ECG). Details from this electrical signal can be used to classify what type of arrhythmia, if any, the patient has by analyzing the PQRST wave properties. The arrhythmias analyzed for this study are Bundle Branch Block, Supraventricular Tachycardia, Ventricular Tachycardia, Tachycardia, and Bradycardia. Multiple data samples of normal ECG characteristics also were read by an artificial neural network (ANN) and analyzed for the differences between a normal signal and an irregular signal. The data was extracted from the MIT-BIH Supraventricular database and the MIT-BIH Arrhythmia database. A typical method used to analyze cardiac arrhythmias is to take the Fast Fourier Transform (FFT) of the signal. In this study, an alternate method is used to predict cardiac arrhythmias. A neural network is designed and programmed with this data and then tested to validate the data set. The pattern recognition tool in MATLAB is then used to analyze and predict the data. The program results were tested to validate the data set. The ANN achieved 98.6% accuracy on the test data. The findings and outcome probabilities from this study are more accurate than some current methods of analysis used today. When neural networks are further used to analyze and test medical data samples, the medical community and patients will experience improvements in the diagnosis of heart abnormalities and early detection of debilitating medical conditions.