Automated Microstress Assessment During Pregnancy Using ECG Sensing and Optimized Deep Networks

Elevated stress levels during pregnancy increase the risk of delivering a premature or low-birthweight infant. Recently, microecological momentary assessment (micro-EMA) has been explored extensively. However, capturing more distinct physiological responses to micro-EMA is still challenging. In this letter, we propose a methodology for micro-EMA-based stress detection using feature extraction and classifiers. For this, an online publicly available micro-EMA database (N=18) is considered. The ECG signals are preprocessed. Ten features are extracted and applied to the classifiers, namely, support vector machine, decision tree, gradient boosting (GradB), adaptive boosting, 1-D convolution network (DL), and, DL with fine-tuning (DLFT). Performance is evaluated using leave-one-subject-out cross-validation. The proposed approach is able to discriminate stress in pregnant mothers. Using DLFT, the approach yields an average classification F1 score, precision, and recall of 76.50 %, 72.40%, and 86.25%, respectively. Thus, the proposed approach could be extended for integrated monitoring systems, enabling real-time stress detection during pregnancy.