Aims: The heart rate (HR) response to paced deep breathing is a common test of autonomic function, where the scoring is based on indices reflecting the overall variability in HR. However, high scores can also be due to arrhythmias. This study presents a method assessing the similarity between HR and breathing based on hyperdimensional computing.
Methods: Data consisted of HR and respiration signals from deep breathing tests (6 breaths/min during one min, corresponding to 0.1 Hz) in 174 healthy subjects and 135 patients with different degrees of autonomic dysfunction and arrhythmias. Three characterizing feature variables were derived from Fourier Series analysis, representing the power in three spectral regions where the power was expected to be low if HR followed the paced breathing pattern. The feature vectors were mapped into hyperdimensional space, using binary basis vectors of dimension N=10000, where approximately the same number of 1’s and 0’s was distributed randomly. The similarity between the hyperdimensional vectors for HR and respiration was assessed using the normalized Hamming distance, which ranged from zero for identical vectors to approximately 0.5 for highly dissimilar vectors. Subjects were divided into four groups based on k-means clustering of Hamming distance. Power spectral percentiles (based on FFT methodology) were determined for each group, showing a clear distinction between different degrees of HR responses, ranging from highly similar to respiration, with a marked peak at 0.1 Hz and its harmonics, to broadband spectral patterns in the lowest similarity class. A comparison was made with a coherence-based index, where similar clustering results were found.
Conclusion: The proposed vector-based similarity analysis is advantageous as it can be based on all types of feature variables, and also presented with equivalent results as the coherence-based analysis, helping to identify abnormal HR responses during deep breathing.