Real-Time Stress Detection Using Single-Lead ECG Signal Analysis

Stress detection is an active research field with ongoing challenges, particularly in real-time data analysis. This pilot study aims to implement a real-time stress detection algorithm that utilizes cardiac and respiratory information extracted from a single-lead ECG signal. Novel methods are combined to extract physiological indices, which are then processed using the XGBoost machine learning technique to provide per-second inferences. ECG signal processing is conducted over a storage buffer to ensure low processing time, with sliding-window analysis applied to extract cardiac indices in the time domain, frequency indices with the orthogonal subspace projection decomposition method, and respiratory indices in both time and frequency domains from ECG-derived respiration. A single-subject experiment, including a relaxation stage, musical stimuli stage, and stressor stage, is conducted to evaluate the algorithm's performance. The comparison between indices extracted via offline and real-time processing yielded a Pearson correlation coefficient average of 0.98 for 21 indices, demonstrating the method's effectiveness. Additionally, the stress probabilities inferred by the model during the experiment clearly distinguish between stages, as supported by visual descriptive statistics and significant differences found in ANOVA and post-hoc tests.