Page 117 - AIH-2-3

Artificial Intelligence in Health Opportunities for AI-based arrhythmia screening

by processing speed, convenience of data routing, and
the implementation of higher mathematical techniques
whenever necessary for complex signal processing.
Examples of signal processing mechanisms are the Fourier
transform, wavelet transform, Wiener filter, Kalman
filter, whitening filter, Butterworth filter, Whiterock
filter, whitening transformation, and matched filter
processing. 19-28 While a majority of these processing tools
are applied in signal denoising techniques, others have
more specific tasks in pattern recognition, particularly
isolated pathological deviations in an individual’s ECG
signal. Analog-based signal processing techniques, such
as frequency cutoffs (low-pass, high-pass, and band-pass
filtering), can be implemented using electronic hardware
and computationally, often in conjunction with Fourier
transform processing. In machine learning applications,
neural networks are used to acquire signals and generate
screening templates that represent various pathological
conditions. However, significant work remains in defining
these templates and customizing them for individual
patients. With more than 50 cardiac conditions to consider,
distinguishing risk factors and electrophysiological
characteristics presents a significant challenge.
Figure 4. Representation of the acquired 12-lead electrocardiogram
recordings: V1, V2, V3, V4, V5, V6, I, II, III, augmented vector right Diagnostic screening is further complicated by the
(aVR), augmented vector left (aVL), augmented vector foot (aVF). stringent confidence levels required for Class IIb medical
The bottom section illustrates the complete lack of signal cohesion in devices. The infrequency of certain arrhythmogenic
ventricular fibrillation. Recording locations: V1: Refer to Figure 3; V2– episodes adds another layer of complexity, necessitating
V4: Record from the front of the heart and are the anterior leads; V5 and
V6: Record from the left side of the heart and are the lateral leads; Lead I careful selection of start and end times for the data stream
(bipolar): Between aVL and aVR, creating a second high lateral lead; Lead used for the creation of analytical templates. Automated
II (bipolar): Between aVR and aVF, generating the second inferior lead; selection of these time frames can be achieved using
Lead III (bipolar): Between aVL and aVF, generating the third inferior neural network processing and other signal-processing
lead; aVF: Left ankle or left lower abdomen; aVL: Left wrist or shoulder; techniques, such as identifying deviations from established
aVR: Right wrist or shoulder.
reference depolarization patterns by omitting peaks
at specific temporal locations. These computational
All data processing for this research is conducted
using high-end personal computers. While some of procedures are integral to the broader toolkit of AI-driven
these analytical procedures can be programmed into neural network processing. In most cases, analog-to-
digital conversion is required before automated signal
existing approved monitoring devices, more complex processing can be performed. Various off-the-shelf analog-
approaches, such as wavelet and matched filter techniques, to-digital converters are available to perform this task
require specialized programming adjustments for use in electronically, enabling digital signal processing. Analog
standalone clinical monitoring devices. Notably, no final signal processing, on the other hand, includes traditional
clinical solutions are provided at this point. techniques, such as signal amplitude control, modulation/

3.1. Programming and its mechanisms demodulation (e.g., frequency modulation and amplitude
modulation), and a plethora of electronic error correction
Various programming languages are used to achieve techniques.
diagnostic modalities and protocols. The input signal is
often managed using various mathematical techniques, 3.2. Signal processing: a theoretical approach to the
including noise reduction, filtering, feature extraction, and determination of depolarization pattern
compression. Examples of the programming languages An important temporal feature in ECG analysis is the
being used are Machine Language, C, C++, Java, Python, duration of the QRS complex, which is typically identified
R, Structured Query Language, Swift, MatLab, and Perl. by its characteristic shape and relatively stable time
The choice of programming language is often determined constant within the waveform. Another critical feature is

Volume 2 Issue 3 (2025) 111 doi: 10.36922/aih.8468