Background: Emotional states like depression increase the mortality and morbidity in heart diseases including inducing arrhythmias and sudden cardiac death. Diagnosis of such emotion-based on the physiological signs can help predict diseases and attendant mortality and morbidity when clinical hallmarks are not apparent.
Objective: To understand if machine learning can achieve the creation of a distinct EKG sequence that can correlate with emotions like depression.
Methods: We extend McSharry et al.’s (IEEE Trans. Biomed. Eng. 2003) dynamical model for synthesizing electrocardiogram (ECG) signals to incorporate the influence of emotions. A mathematical model of heart rate (HR) kinetics is used to generate RR-interval time-series in response to different activities and emotions. To incorporate emotions, a machine learning model is used to learn a mapping from {valence, arousal} to changes in HR and HR variability from data of 62 subjects obtained from benchmark datasets (DEAP, HCI). The outputs of the model are added to HR demand. The final output, instantaneous HR time-series, is converted to RR-interval time-series which is applied to McSharry et al.’s model for generating ECG.
Results: The proposed model is evaluated for effect of emotions including depression by comparing the direction of change in each synthesized signal with the direction of change in the corresponding signal in Kreibig’s (Biol. Psych. 2010) table. The synthesized signals yield an accuracy of 96.7%. The HR and respiration rate synthesized as a function of different action intensities had significant correlation with the corresponding signals from each of 22 subjects in benchmark datasets.
Conclusion: To our knowledge this is the first time, a computational model is presented here to synthesize the ECG signal as a function of emotions. Experimental results show high fidelity between the synthesized signals and the benchmark data under comparable conditions. Direct comparison of patient's data with such synthesized EKG can unravel subclinical conditions that effect mortality and morbidity.