Page 112 - AIH-2-4

Artificial Intelligence in Health Autonomic nervous system patterns in men

referred to as classification analysis or numerical taxonomy, scale (Filizola, Brazil). All anthropometric measurements
cluster analysis differs fundamentally from discriminant were performed by a trained assessor to minimize inter-
analysis, which requires predefined groups. In contrast, rater variability and ensure data accuracy.
cluster analysis aims to identify previously unknown
groupings inherent in the data. 42 3.3. Experimental procedure and data acquisition

The cluster analysis process typically involves a series The tests were conducted in a quiet room maintained at
of key steps: defining the research problem; selecting an a temperature of 22°C. Participants were instructed to
appropriate distance or similarity measure; choosing a refrain from strenuous physical activity for 24 h and to
clustering algorithm; determining the optimal number of avoid consuming alcohol, caffeine, or large meals for at
clusters; interpreting the characteristics of each identified least 3 h before their session. Upon arrival at the laboratory,
cluster; and evaluating the validity of the resulting cluster participants rested quietly in a supine position for 10 min
solution. Careful selection of variables is essential and while breathing spontaneously. RRIs were recorded
43
should be guided by research hypotheses, prior studies, throughout this period using a Polar V800 heart rate
and the researcher’s informed judgment. Similarly, the monitor (Polar, Finland) with a sampling rate of 1,000 Hz.
choice of distance or similarity measure is critical; for The monitor was positioned over the xiphoid process of the
instance, Euclidean distance is frequently used. 44 sternum. The first 5 min of data were discarded to allow for
signal stabilization, and the subsequent 5 min were used
Clustering methods are broadly classified as for analysis. The tachograms of RRI were transferred via
hierarchical, non-hierarchical, or two-stage. Hierarchical an infrared interface to Polar Precision Performance SW
43
approaches build a nested structure of clusters either software version 3.0 (Polar, Finland), which automatically
agglomeratively (bottom-up) or divisively (top-down) corrected the RRI using a moving average filter. The data
and do not require a predefined number of clusters. The were then saved as “.txt” files.
results of these methods are visualized using dendrograms,
in which branch lengths indicate inter-cluster distances. 3.4. HRV analysis
In contrast, non-hierarchical methods such as K-means For the time-domain analysis, the following parameters
require the number of clusters to be defined in advance. were calculated: MRR, SDNN, RMSSD, and the pNN50.
The choice of method depends on the distance measure For the frequency-domain analysis, spectral analysis was
used, and the resulting clusters must be interpretable and performed using the Welch periodogram method (256-point
relevant to the research objectives. segments, 128-point overlap, and a Hanning window). This

3. Materials and methods yielded normalized power for the LF (0.04 – 0.15 Hz) and
HF (0.15 – 0.40 Hz) bands, both expressed as percentages.
3.1. Study population All parameters were computed in accordance with the
This cross-sectional study was conducted in Macapá, guidelines established by the Task Force of the European
Brazil, and involved 80 healthy, young male participants Society of Cardiology and the North American Society of
24
(22.0 ± 2.8 years). Participants were recruited based on Pacing and Electrophysiology, and were implemented in
a low-risk profile for CVD. Exclusion criteria included MATLAB 2020.b (MathWorks, United States).
45
smoking, a history of cardiopulmonary disease, or the 3.5. Statistical analysis
current use of any medication. All participants provided
verbal informed consent before enrollment. The study Descriptive statistics are presented as mean ± standard
protocol was approved by the Human Research Ethics deviation. The Shapiro–Wilk test was employed to assess
Committee of the Federal University of Amapá (CAAE: the normality of the data distribution.
50150121.1.0000.0003) and conducted in accordance An 80 × 6 matrix of normalized HRV data derived from
with the principles of the Declaration of Helsinki and the RRI tachograms was used for dimensionality reduction
46
Resolution 510/2016 of the National Health Council. to two dimensions using PCA. PCA, a dimensionality
reduction technique, transforms correlated variables into
3.2. Anthropometric assessment
uncorrelated PCs via eigenvalue decomposition of the
Before enrollment, all participants received a detailed covariance matrix. K-means clustering an unsupervised
15
explanation of the study protocol, including measurement learning algorithm – was then applied to the reduced-
procedures and estimated duration. Participants were dimensionality data, making it well-suited to classify
16
instructed to wear appropriate attire (e.g., light clothing, individuals into distinct groups based on their ANS
no shoes) and to avoid carrying objects. Height (cm) and regulation. Cluster assignment was based on the Euclidean
weight (kg) were measured using a calibrated mechanical distance metric (Equation I). Cluster centroids were

Volume 2 Issue 4 (2025) 106 doi: 10.36922/AIH025050006