Acute mental stress level detection: ECG-scalogram based attentive convolutional network

Abstract

In today’s competitive environment, everyone is under psychological stress. Long-term exposure to stress can lead to serious issues such as high blood pressure, depression, violence, cardiac and brain damage, and even suicide. To live a healthy lifestyle, it is critical to monitor stress and its levels regularly. Existing methods detect stress specifically, whereas detection of multiple levels of stress has yet to be explored. To address this issue, the paper presents the lightweight ECG-stress-ScaloNet model, which employs an attentive convolutional neural network (CNN) to analyze short-term ECG scalogram images. In this work, a unique inception-attention block is created. The inception module captures multi-scale information; additionally, attention focuses on extracting meaningful features from multi-scale feature maps by utilizing cross-channel and spatial information. Two databases are used to evaluate the proposed ECG-stress-ScaloNet model. The first is Physionet driver stress and normal ECG data that is publicly available, and the second is self-created academic practical-viva stress and normal ECG data. The ECG-stress-ScaloNet outperforms the existing methods, with a test accuracy of 98.28% for the Physionet dataset and 95.71% for the self-created dataset. For the intended application, the ECG-stress-ScaloNet model is reliable and accurate since it has fewer learnable parameters and decreases computational complexity.