محمد حسن زاده سورشجاني

Electrocardiography is a diagnostic tool that records the electrical activity of the heart. ECG signals can indicate various heart abnormalities and diseases, such as Covid-19, a novel coronavirus disease that can cause severe respiratory and cardiovascular complications. In this thesis, we used convolutional neural networks (CNN) to classify ECG signals of patients with covid-19 and other cardiac abnormalities, such as infarction and history of infarction. We used data augmentation techniques to increase the number and diversity of ECG signals and used feature extraction to improve the performance of our CNN model. Before applying the CNN model to the ECG signals, we also performed a feature extraction method to convert the ECG signals into spectrograms. Spectrograms are visual representations of the frequency and amplitude of ECG signals over time. We used short-time Fourier transform (STFT) to calculate spectrograms from ECG signals. STFT is a technique that divides ECG signals into short segments and applies a Fourier transform to each segment. This allowed us to record both the time and frequency information of the ECG signals. We eva‎luated our model using precision, accuracy, recall, and F1 precision as benchmarks and compared it with a baseline model that uses hand-crafted features. Our preliminary results accurately show that our CNN model performs well in binary and multi-class classification tasks, achieving a high average accuracy of 92.48 and an F1 score of 90.96. We believe that our project has been able to provide a quick and accurate method to identify and monitor various cardiac conditions using ECG analysis

الكتروكارديوگراف , كوويد , شبكه‌هاي عصبي كانولوشن , عددي سازي سيگنال

ECG , COVID , Convolutional Neural Network , digitizing signal

Mohammad Hassanzade

Dr. Javad Vahidi