Atrial Fibrillation Automatic Diagnosis Based on ECG Signal Using Pretrained Deep Convolution Neural Network and SVM Multiclass Model

The paper presents a robust deep learning approach for ECG automatic diagnose. For this purpose, Deep Convolution Neural Network (D-CNN) algorithm and a multiclass model for SVM classifier will automate the detection process of ECG images specific to atrial fibrillation cases. In this research work, a pre-built and pre-trained D-CNN model is developed. It applies transfer learning which has been proved as a robust technique for computer vision. The early layers of convolutional network are frozen and only the last few layers are trained, identifying objects in images either through a database search or through real-time analysis and detection of the fetched image. Further, the study includes a comparison between the results of using data augmentation techniques and the results without using it. We achieved an average 99.21% of accuracy. The implementation environment of our work is based on MATLAB using Deep Network Designer toolbox.