ECG Quality Assessment Using Deep Learning

Human body is the source of many electrical signals that give information about its physical condition. These signals carry an information that can be processed automatically and used in countless applications. However, for this to be possible, the signal must be of sufficiently high quality. In this work we propose a deep learning-based method for evaluation the quality of ECG signal, which is the most commonly used signal for examination of the heart condition. For this purpose we used publicly available database of single-lead ECG signals. The data were processed using well-known algorithms to obtain several hand-crafted features that were then used to train a feedforward neural network. The model yielded accuracy, precision, and recall values above 81%, and area under ROC value of 96.4%. The created method can be implemented directly on hardware which makes it even more desirable.