Page 119 - AIH-2-3

Artificial Intelligence in Health Opportunities for AI-based arrhythmia screening

Table 1. Overview of common temporal and frequency and repolarization (on- and off-ramps) can be altered
analysis tools under pathological conditions, requiring an extensive
analytical frequency bandwidth to accurately capture these
Time domain Frequency domain
variations in the frequency domain. 8
Analog-to-digital conversion Adaptive filtering
Amplitude Band-pass filtering In the Fourier domain, a standard (healthy) ECG can
Autoregressive model Continuous wavelet transform be accurately described using the first eight harmonics
of the heart rate. However, even minor high-frequency
Bilinear transform Convolution deviations from the conventional ECG waveform often
Continuous Correlation introduce alterations, requiring a larger number of
Convolution Discrete cosine transform higher harmonics to characterize the frequency-domain
Correlation Discrete Fourier transform features. The baseline ECG typically has a root frequency
Discrete Empirical mode decomposition of approximately 1 Hz, with higher harmonics extending
Discriminant functions (least Fast Fourier transform up to 8 Hz for root analysis. As a general guideline,
squares/synthetic) frequency analysis of an ECG should at least cover a
Duration Finite impulse response filtering range of 0.0001–100 Hz for a normal ECG, though the
Event analysis (missing Fourier transform fundamental frequency may be excluded from the analysis.
events) For arrhythmogenic tendencies, spectral analysis, such as
Finite impulse response High-pass filtering the Fourier transform, may need to extend beyond 200 Hz
filtering to capture relevant features. Further analysis is needed to
Infinite impulse response Infinite impulse response filtering determine the type of arrhythmias, requiring advanced
filtering mathematical methods to investigate the isolation and
Interval Inverse Fourier transform confidence levels of the specific options identified for the
Least squares discriminant Laplace transform final diagnosis. Nonetheless, the final determination of
functions the applicable arrhythmogenic pathology may still rest
Least squares method (in Least-squares discriminant analysis with the physician. It is important to note that extending
time series) the analysis to even higher frequencies is unlikely to yield
Matched filter technique Least-squares wavelet analysis additional diagnostic information, as the spectrum beyond
Recurring sequence of events Low-pass filtering this point is typically dominated by noise.
Sampling property of unit Matched filter techniques The signal processing techniques discussed here
impulse are examples of early use of AI in medical diagnostics,
Synthetic discriminant Phase-shift analysis particularly in cardiac rhythm analysis, risk assessment,
functions and pathological conditions screening. With advancements
Time-reversal technique Sampling (e.g., Nyquist–Shannon in knowledge, wavelet analysis has become increasingly
sampling theorem), analog-to-digital utilized, while the matched filter approach is gaining
conversion prominence in analyzing more complex aspects of cardiac
Trigger point Spectral confinement depolarization pathology.
Wavelet transform Spectral decomposition
Z-transform Synthetic discriminant functions 3.4. Wavelet transform analysis
S-transform Since action potentials are mainly stochastic, wavelet
Wavelet transform analysis may not always provide the volume or quality
Windowing of data required for precise diagnosis. However, when
Z-transform analyzing a repetitive signal as the vector summation of
Notes: Some specific frequency filters: Wiener filter, Kalman filter, multiple action potentials, wavelet-domain features can
effectively classify the relative contributions of higher
whitening filter, Butterworth filter, Whiterock filter, and whitening
transformation; All diagnostic modalities are represented in no harmonics. The wavelet features used in the analysis of
11
particular order, alphabetically ranked. an ECG signal mainly detect scaled or shifted versions
of typical patterns or the aspects of depolarization wave.
exceeding 200 beats/min. It is worth noting that a teenager Wavelet decomposition using a mother wavelet that
engaged in sporting activities may easily reach or exceed closely resembles the general shape of the QRS complex
a heart rate of 240 beats/min without immediate health (i.e., baseline template: generic QRS complex), regardless
concerns. Beyond differences in heart rate, depolarization of electrode numbering and placement, enables the

Volume 2 Issue 3 (2025) 113 doi: 10.36922/aih.8468