Advances in high resolution joint time-frequency biosignal analysis and visualization can transform autonomic function assessment

      Background: Over the past decades, autonomic nervous system function estimation based on the analysis of variability in heart rate (HRV), blood pressure and related cardiovascular and neurophysiological signals has mainly employed the derivation of parameters from Fast Fourier Transform (FFT) or autoregressive modeling spectral analysis. While this approach has become canonical and popular, it suffers from a number of severe limitations. Among those is the FFT prerequisite of a stationary signal composed of sinusoid oscillations, as well as the low resolution time-averaged results that obscure transient or event-specific features of the signals. Similarly, multimodal signal analysis has been mainly constrained to coherence and transfer function types of analysis. Aim: Evaluation of the utility of recently available very high resolution time-frequency signal decomposition techniques for HRV and related biosignals in experimental paradigms relevant to autonomic nervous system research. Methods: In addition to conventional HRV analysis methods and those proposed in recent years to partially overcome the shortcomings of FFT (such as short-time FFT, Wigner-Ville transform and others), we implemented advanced wavelet transform algorithms with highly parallelized computing acceleration, specially designed wavelets and secondary wavelet feature decoupling and comparatively applied these techniques to several hundred recordings with signals of various quality and from diverse sources such as rodent and human neonate, child and adult, different experimental paradigms, including highly dynamic processes difficult for traditional analysis, such as tilt table tests. Result: While new, arbitrarily high resolution joint time-frequency analysis techniques appear to open a new window for precise interrogation of dynamic events in cardiovascular autonomic data, they demonstrate remarkable robustness against signal artefacts, such as misclassified or missing beats in HRV, requiring markedly reduced manual beat time series cleanup effort. At the same time, it is still possible to relate to established autonomic function estimation concepts like high and low frequency components of HRV, but those can now accurately localized in time, quantified and characterized in their dynamic change. We also demonstrate that this approach benefits multimodal signal analysis by enabling the search and identification of complex patterns in signals from different physiological sources at varying time scales, including imaging. Conclusions: With the growing availability of high performance computing resources, the powerful signal analysis methods described here can be routinely applied in autonomic research, providing the ability to search for and identify events and patterns in biosignals at arbitrary timescales and precision and in multimodal signal acquisition and imaging experimental paradigms, breaking through the limitations of traditional methology. Furthermore, they can be employed as signal preprocessing and feature extraction steps for further analysis with supervised and unsupervised machine learning technologies for complex pattern detection and classification.
      To read this article in full you will need to make a payment


      Subscribe to Autonomic Neuroscience: Basic and Clinical
      Already a print subscriber? Claim online access
      Already an online subscriber? Sign in
      Institutional Access: Sign in to ScienceDirect