کنفرانس مهندسی برق ایران

صفحه اصلی / سی و دومین کنفرانس بین المللی مهندسی برق

Using Convolutional Neural Networks for Sudden Cardiac Death prediction

نویسندگان :

Sara Tavazo¹ Farideh Ebrahimi²

1- دانشگاه صنعتی نوشیروانی بابل 2- دانشگاه صنعتی نوشیروانی بابل

کلمات کلیدی :

Electrocardiogram (ECG)،(Sudden Cardiac Death (SCD،(Convolutional neural networks (CNN،( Continuous Wavelet Transform (CWT

چکیده :

The purpose of this study was to predict Sudden Cardiac Death early to treat cardiac disorders effectively and reduce mortality caused by a delayed diagnosis. Traditional methods have relied on analyzing Electrocardiogram and Heart Rate Variability signals for SCD prediction; their success, however, heavily depends on the feature extraction process. Therefore, Convolutional Neural Networks seem to be a suitable alternative for automatic feature extraction. This study, for the first time, presents a method for predicting SCD 60 minutes before it occurs using one-dimensional and two-dimensional CNNs. At first, a Model based on the One-Dimensional Convolutional Neural Network was used for SCD prediction, and by the proper setting of parameters, an accuracy of %98.6 was obtained. Then, due to the success of CNNs in image analysis, the ECG signal was converted to Two-Dimensional images to be used as input in 2-Dimensional Convolutional Neural Network, which by applying proposed architecture, the classification accuracy increased to 99%. Finally, in order to reduce complexity, some changes were made in the 2D-CNN based proposed algorithms. These changes include reducing the number of filters, reducing the number of final parameters of the network by adding a global average-pooling layer before the fully connected layer, and adding one more convolution layer to preserve the efficiency of the network. After applying these changes, the accuracy was %98.68 in SCD prediction.In addition to being simple and effective, the methods proposed in this research provide the highest accuracy and maximum prediction time.