A novel method for distinction heart rate variability during meditation using LSTM recurrent neural networks based on visibility graph

Detection of dynamical complexity variations in heart rate signals during meditation practice can be considered as an important criterion for further understanding cardiovascular functions. This paper introduces a computational framework for categorizing the heart rate variability (HRV) differences of subjects during meditation compared to before meditation using information from heart rate signals. Heart rate data are obtained from the Physionet database. Because of the dynamic and nonlinear nature of heart rate signals, visibility graph features were utilized for analysis. The recurrent neural network (RNN) based on long short-term memory (LSTM) was used to classify subjects into two different states before and during meditation relying on its performance in terms of the VG-based features. The results show that the proposed approach managed to categorize the heart rate signals into two groups: before and during meditation, with a high accuracy of 99.25%. Moreover, the proposed approach enjoys the potential to extremely ease the adoption of the running situation of the cardiac system and physiological functions by disclosing remarkable differences between the HRV reactions of the two mentioned states.