Detection of Panic Attacks Using Convolutional Neural Networks on ECG Data

Panic attacks are a prevalent psychiatric disease characterized by abrupt and acute episodes of panic, accompanied by specific physical symptoms such as heart palpitations and dyspnea. The prompt and precise identification of these episodes is crucial for enhancing patients’ quality of life. This study proposes an artificial intelligence-based classification method based on electrocardiography (ECG) signals to detect panic attacks. Raw ECG signals are visualized by obtaining two-dimensional scaled representations (scalograms) using continuous wavelet transform (CWT). Then, these images are classified with deep learning-based convolutional neural network (CNN) models. In the study, scalogram images were obtained using 5-second ECG signal segments. Then, they were classified with various CNN models, and their performances were evaluated. The findings indicated that the ResNet50 model exhibited the superior classification performance, achieving an accuracy of 99.83% and an AUC value of 100%. Consequently, it demonstrates that the utilization of time-frequency-based information processed using deep learning models yields great accuracy in detecting panic attacks.