Page 111 - AIH-2-4

Artificial Intelligence in Health Autonomic nervous system patterns in men

reliability of these metrics is highly dependent on the averaging (PRSA), a statistical technique designed to
integrity of the RRI data, as artifacts such as ectopic beats analyze quasi-periodic signals in non-stationary or noisy
can significantly distort the results. Consequently, the data. By applying PRSA to NN intervals, they proposed
application of filtering techniques is a crucial preprocessing deceleration capacity (DC) and acceleration capacity
step. A commonly used approach involves excluding any metrics derived from the coherent averaging of RRIs that
RRI that deviates by more than 20% from the preceding exhibit increases or decreases, respectively. These indices
normal interval, thereby ensuring data accuracy. 25 aim to assess sympathetic modulation of sinoatrial node

Spectral analysis of HRV quantifies the power acceleration and deceleration, independent of other
distribution of different frequency components within the physiological factors.
sinus rhythm. The two primary methods employed are the The DC index, in particular, has gained significant
non-parametric Fourier transform which decomposes the attention due to its promising clinical implications.
signal into constituent sinusoids – and the parametric Notably, studies have demonstrated its superior predictive
26
autoregressive model, which estimates the spectrum power for mortality following acute myocardial infarction
using a predictive model of the RRI. Despite their compared to the widely used left ventricular ejection
27
34
methodological differences, both methods provide broadly fraction. Furthermore, a strong relationship has been
comparable assessments of HRV spectra, and neither observed between DC and the risk of sudden cardiac death
35
demonstrates a clear advantage over the other. in individuals with Chagas disease. Studies have also
reported a significant correlation between DC and high
Spectral HRV analysis typically distinguishes two 36,37
main frequency bands: the high-frequency (HF) band levels of physical conditioning, suggesting its potential
use as a valuable marker of cardiovascular health and
(0.15 – 0.40 Hz), which reflects parasympathetic modulation fitness.
associated with respiratory sinus arrhythmia, and the low-
frequency (LF) band (0.04 – 0.15 Hz), which represents 2.2. PCA
a combination of sympathetic and parasympathetic PCA is a dimensionality reduction technique that
influences on baroreflex regulation. The LF/HF ratio is transforms a set of correlated variables into a smaller
28
often calculated to estimate sympathovagal balance, with the number of uncorrelated linear combinations, known
HF band serving as an index of parasympathetic tone and as principal components (PCs). These components are
the LF band representing integrated autonomic output. 29
ordered to capture as much of the total variance in the
Similar to the time-domain analysis, the accuracy of original dataset as possible. 38
HRV spectral analysis is highly dependent on data quality The first PC captures the largest proportion of the total
and requires careful handling of arrhythmias. A common variance. The second PC explains the maximum remaining
approach is to exclude RRIs immediately before and after variance, with the constraint that it is uncorrelated with
ectopic beats and replace them with interpolated values the first. This process continues sequentially, with each
based on adjacent, true RRIs. However, the exclusion subsequent component capturing a decreasing proportion
of more than two RRIs surrounding an ectopic beat is of the remaining variance and remaining uncorrelated
generally avoided due to the risk of compromising signal with all previously derived components. 39
continuity. 30
Although PCA can theoretically continue until all
Following the Task Force report, various non-linear variance is accounted for, it is typically stopped after
24
metrics have emerged to analyze the complex, multi-causal, extracting a smaller number of PCs that collectively
and potentially chaotic nature of HRV. These metrics explain a significant proportion of the total variance. The
38
apply techniques such as Lyapunov and Hurst exponents, eigenvalue associated with each PC represents the amount
coarse-grained spectral analysis, detrended fluctuation of variance it explains higher eigenvalues indicate greater
analysis, and entropy measures to capture the interplay explanatory power. 40
of humoral, hemodynamic, and electrophysiological
factors influencing HRV. While their precise physiological 2.3. Cluster analysis
interpretations remain under investigation, these methods Cluster analysis encompasses a range of statistical
have shown promising potential in differentiating the techniques used to group an initially unclassified set of
effects of conditions such as stress and diabetes on HRV. 31,32
cases, subjects, or objects into relatively homogeneous
Novel metrics based on instantaneous heart rate groups, or clusters, based on observed characteristics. 41
acceleration and deceleration have also been introduced. The primary goal is to identify underlying group structures
In 2006, Bauer et al. developed phase-rectified signal without prior knowledge of group membership. Also
33

Volume 2 Issue 4 (2025) 105 doi: 10.36922/AIH025050006