A deep neural network to identify irregular cardiac rhythms from single lead ECG signals at a significant performance and diagnostic yield is being developed which can reduce the load of cardiologists. Our neural network can identify up to 7 arrhythmias along with sinus rhythm. It is correlated to with clinically validated reports to enhance the output.
The network developed is a convolutional ANN which takes raw ECG signals as input that are sampled at a frequency of 250 Hz (250 samples per second). The network takes only ECG signals as input that are pre processed before fed to the network. This architecture contains 4 layers. The dataset contains over a lakh of patients > 18 years of age from various demographic ratios, whose cardiac rhythms have been obtained using Biocalculus.