Heart Response to Harmonic Music Interval Stimuli Via Deep Learning Structures

Abstract

The effect of music on the heart is reflected in variables such as heart rate and electrocardiographic (ECG) signals. ECG is a record of heart electrical activity and is a useful tool in diagnosing various cardiopathies. Artificial intelligence techniques have recently been implemented to analyze ECG and RR-interval data and are used thus in the present study to examine the influence on the heart of harmonic musical intervals and colored noise. Harmonic intervals were chosen because of their emotional response, while noise has been linked to positive responses such as improved sleep quality. A deep learning system was implemented, employing the ResNet-18 and GoogLeNet pre-trained networks to discriminate 31 different classes of ECG and RR-interval responses to the sound stimuli. Following an exploratory approach, deep learning was selected as an alternative to traditional analysis with the expectation that it could be incorporated into future music perception research. Classification revealed the ability of the implemented system to demonstrate heart response to the stimuli. ECG signals performed best, with 97% accuracy and Matthew's coefficient of 0.97, while RR-interval achieved a 93% accuracy and Matthews coefficient of 0.93, suggesting that the considered stimuli of harmonic musical intervals and noise produced different responses in the heart. Moreover, the Matthews coefficient values above 0.7 and close to 1 imply a correlation between the two types of stimuli and the heart response, as measured by ECG and RR-interval signals.