Review of ECG Telemonitoring Techniques Using

Neural Network-Based Feature Extraction

Arshad Abhineet Mishra Abhishek Kumar
Kiet Group of Institution Kiet Group of Institution Kiet Group of Institution
Department Of Electrical and Department of Electrical and Department of Electrical and
Electronics Engineering Electronics Engineering Electronics Engineering
arshad.2226en1010@kiet.edu abhineet.2226en1003@kiet.edu abhishek.2226en1032@kiet.edu

Supervisor
Dr. Snigdha Chaturvedi
Kiet group of institution
Department of Electrical and
Electronics Engineering

Abstract- Electrocardiography (ECG) is a vital, non-invasive Early detection and intervention are crucial for improving patient
method for assessing heart function and diagnosing cardiac outcomes [7]. Traditional methods for ECG analysis often
issues. Recent research has significantly advanced heart require lengthy hospital stays or clinic visits, limiting their
disease prediction using ECG data through machine learning accessibility [6, 10].Wearable wireless sensor networks
and deep learning approaches. Several studies developed (WWSNs) offer a promising solution for continuous and remote
accurate ECG-AI models using 12- and single-lead ECG ECG telemonitoring [4, 10]. These networks consist of wearable
data, achieving high accuracy in predicting fatal coronary sensors that capture ECG data as shown in Figure 1 and transmit
heart disease (FCHD) risk when incorporating demographic it wirelessly to a central processing unit [4]. Machine learning,
and clinical data. Benchmarking revealed that convolutional particularly deep learning, has emerged as a powerful tool for
neural networks (CNNs), particularly ResNet and inception analyzing ECG signals [6, 9]. Deep learning models can
architectures, are highly effective for ECG analysis. Various automatically extract features from the ECG data, eliminating
innovative methods, such as decision tree-based random the need for manual feature engineering and potentially
forest classifiers, wavelet transform, and independent improving accuracy [6, 9].This research investigates the
component analysis, have shown improvements in heart application of a WWSN for ECG telemonitoring, with a focus on
disease and arrhythmia classification. Transfer learning has utilizing a neural network for feature extraction from the ECG
also proven beneficial, as demonstrated by a CNN model for signal. This approach has the potential to improve accessibility,
heartbeat classification applied across different ECG convenience and potentially diagnostic accuracy for ECG
conditions. Additionally, explainable AI techniques have analysis, aiding in early detection and management of heart
ensured model transparency and reliability in predicting conditions.
heart disease. The use of IoT and wireless sensor networks
for ECG telemonitoring has proven effective for remote
monitoring of heart rate. An advanced model combining
CNNs and RNNs has shown high sensitivity and specificity in
detecting atrial fibrillation. Machine learning applied to
electronic health records has improved pre-emotive
cardiovascular disease risk assessment. Overall, these studies
highlight the potential of combining wearable wireless
sensors with neural networks for real-time, remote cardiac
health management

keywords-ECG(electrocardiogram), MLP(multi-layer-
perceptron), KAN(Kolmogorov-Arnold networks), Neural
Networks.

I. INTRODUCTION

Summary Introduction: Wearable Wireless Sensor Networks for

ECG Telemonitoring with Neural Network Feature Extraction
Atrial Fibrillation (AF) and cardiovascular disease (CVD) are
major health concerns with significant economic burdens [1, 7]. Figure 1 Ecg Synthetic Data
 II. LITERATURE REVIEW

In recent years, the integration of advanced technologies has emphasizes the significance of visualization methods for
greatly transformed healthcare, improving the ways we monitor, detecting anomalies. Additionally, it explores advancements in
diagnose, and manage various medical conditions. Notably, wireless technologies and discusses the advantages of real-time
wearable wireless sensor networks combined with machine monitoring and wearable ECG systems. The paper also addresses
learning, particularly neural networks, have significantly key challenges such as motion artifacts and signal noise, while
enhanced electrocardiogram (ECG) telemonitoring systems. This emphasizing how wearable sensors and intelligent healthcare
paper offers a thorough review of modern ECG telemonitoring solutions are driving significant innovations in ECG monitoring.
techniques that utilize neural network-based feature extraction, At last, this paper also concludes that more advancement in ECG
highlighting the substantial advancements in early detection and monitoring can be done by adding more sensors that can track
management of cardiovascular diseases. The subsequent sections the patient’s movements and can predict any abnormality in real
discuss the methods and approaches currently in use in this field. time.

1: Transfer Learning for ECG Analysis[1] 3: Kolmogorov-Arnold Networks (KANs) for Function
Approximation[3]
This research paper discusses the performance of models trained
on the PTB-XL dataset and tested on the ICBEB2018 dataset for
ECG analysis. It highlights the effectiveness of transfer learning
from a larger dataset to a smaller one, with the xresnet1d101
model showing the best performance. The results emphasize the
importance of deep learning algorithms, specifically
convolutional neural networks, in ECG analysis. Models trained
on PTB-XL and fine-tuned on ICBEB2018 outperform those
trained from scratch on ICBEB2018. The xresnet1d101 model
demonstrates robustness and adaptability across different
datasets. Deep learning algorithms, especially CNNs, perform
well in ECG analysis tasks. Large, diverse datasets like PTB-XL
enable comprehensive benchmarking and evaluation of
algorithms. Paper 1: Transfer Learning for ECG Analysis

This research paper discusses the performance of models trained

on the PTB-XL dataset and tested on the ICBEB2018 dataset for
ECG analysis. It highlights the effectiveness of transfer learning
from a larger dataset to a smaller one, with the xresnet1d101
model showing the best performance. The results emphasize the
importance of deep learning algorithms, specifically
convolutional neural networks, in ECG analysis. Models trained
on PTB-XL and fine-tuned on ICBEB2018 outperform those
trained from scratch on ICBEB2018. The xresnet1d101 model
demonstrates robustness and adaptability across different
datasets. Deep learning algorithms, especially CNNs, perform
well in ECG analysis tasks. Large, diverse datasets like PTB-XL
enable comprehensive benchmarking and evaluation of
algorithms.

2: Comprehensive Review of ECG Monitoring Systems[2]

This paper provides a comprehensive review of ECG monitoring

systems, covering their architecture, operational processes,
challenges, and emerging trends. It highlights the critical role of
ECG in diagnosing cardiovascular diseases and proposes a
taxonomy of monitoring systems along with an architectural
model. The integration of advanced technologies such as AI and
IoT is essential for enhancing the effectiveness of healthcare
monitoring. The study investigates various machine learning
algorithms used in IoT cloud-based monitoring systems and
In this paper, Kolmogorov-Arnold Networks (KANs) in Figure 2 components: ECG sensing nodes for signal acquisition, Gateway
are proposed as promising alternatives to multi-layer perceptrons nodes equipped with Multilayer Perceptron Neural Networks
(MLPs), which are fundamental to deep learning due to their (MLPNN) for noise reduction and feature extraction, and an IoT
ability to approximate nonlinear functions as supported by the platform enabling seamless data transmission and remote access.
universal approximation theorem. Despite their widespread use, This architecture not only provides immediate access to ECG
MLPs have significant drawbacks, such as high parameter data for healthcare professionals but also empowers patients with
consumption and low interpretability, particularly in transformer self-monitoring capabilities, potentially reducing hospital visits
models. KANs, inspired by the Kolmogorov-Arnold and enhancing healthcare efficiency. To validate the
representation theorem, differ from MLPs by utilizing learnable effectiveness of MLPNN in accurately extracting R-peaks and
activation functions on edges (weights) instead of fixed functions calculating heart rates from raw ECG signals, the study leverages
on nodes (neurons), replacing linear weight matrices with datasets from the MIT-BIH database. Experimental results
learnable 1D functions parameterized as splines, which reduces demonstrate robust performance in various environments,
computational complexity. This hybrid structure leverages the including indoor and outdoor settings, with ECG sensing nodes
strengths of both splines and MLPs, addressing the curse of operating reliably on battery power over extended periods.
dimensionality by splines and the lower accuracy of MLPs in Technical evaluations include rigorous assessments of system
low dimensions. Empirically, KANs demonstrate superior parameters such as autocorrelation, end-to-end delay, and
performance in function fitting and small-scale AI and science coverage, confirming the system's reliability and feasibility for
tasks, achieving comparable or better accuracy than larger MLPs real-world deployment. Furthermore, the research highlights the
and exhibiting faster neural scaling laws. Applications of KANs potential scalability of the framework by suggesting future
include solving physical equations, partial differential equations enhancements, such as integrating additional sensor nodes to
(PDEs), and integration into machine learning models such as monitor additional vital signs and optimizing WSN parameters
transformers, resulting in innovative models like "transformers". for improved efficiency and adaptability. In conclusion, this
The interpretability of KANs makes them valuable for scientific research paper provides a thorough analysis of wearable WSN
collaboration, aiding in the rediscovery of mathematical and and IoT integration for ECG telemonitoring, showcasing
physical laws. Despite limited mathematical understanding, advancements in remote healthcare monitoring systems. By
KANs show promise for enhancing accuracy, efficiency, and bridging the gap between continuous health data acquisition and
interpretability in various AI and scientific applications, timely medical intervention, the study contributes to
positioning them as viable alternatives to MLPs in the evolving personalized healthcare delivery and improved patient outcomes.
landscape of deep learning. Hence, this research is very recent, Future directions include refining the system architecture and
and there has been nearly no research done on ECG feature expanding its application to include a wider range of
extraction using Kolmogorov-Arnold networks. Further research physiological parameters, further advancing remote health
can be conducted to explore the application for arrhythmia monitoring capabilities in clinical practice.
detection and other cardiovascular diseases.
n
z j = ∑ wij x i +b j
i=1

5: HDP-DTRF Methodology for Heart Disease

Prediction[5]

The article explores the early prediction of heart disease through

supervised learning, specifically utilizing stochastic gradient
boosting. It introduces the HDP-DTRF methodology, which
leverages a decision tree-based random forest classifier to
enhance prediction accuracy. This approach aims to improve the
Figure 2 MLP VS KAN
precision of heart disease predictions using advanced machine-
learning techniques. The DTRF classifier, an ensemble learning
4: Integration of Wearable Wireless Sensor Networks model, is particularly effective in managing noisy data and
(WSN) with IoT for ECG Telemonitoring[4] mitigating overfitting issues. Its adaptability and parameter-free
nature significantly contribute to its robust performance. By
This research paper investigates the integration of wearable integrating multiple decision trees, the DTRF model generates
wireless sensor networks (WSN) with the Internet of Things more accurate classification outcomes. The proposed HDP-
(IoT) for ECG telemonitoring, as explored by Amina El Attaoui DTRF approach demonstrates notable improvements in
et al. It addresses a critical healthcare need for efficient remote precision, recall, F1-score, and overall accuracy compared to
monitoring solutions, particularly beneficial in managing chronic traditional methods, marking a significant advancement in the
diseases and improving the quality of life for elderly populations. early detection of heart disease. This HDP-DTRF approach
The focus on real-time ECG signal monitoring underscores the resulted in 86% of precision, 86% of recall, 85% of F1-score,
importance of continuous health data acquisition in the early and 96% of accuracy on publicly available real-world datasets,
detection and proactive management of cardiac abnormalities. which are higher than traditional methods.
The proposed system architecture consists of three main
 M 8: ECG-AI Models for Fatal Coronary Heart Disease (FCHD)
F(x) = ∑ γm hm (x ) Prediction[8]
m=1
This study aimed to create precise ECG-AI models for predicting
6: Deep Learning Model for Real-time Atrial Fibrillation the risk of fatal coronary heart disease (FCHD) and assessing
(AF) Detection[6] time-dependent risk factors. By analyzing ECG data from over
50,000 patients, the researchers found that the most effective
The study presents an innovative deep-learning model designed model combined demographics, clinical data, and ECG-AI,
for the real-time detection of Atrial Fibrillation (AF) in ECG achieving high accuracy in predicting FCHD risk. Both the 12-
recordings. This model integrates Convolutional Neural lead and 1-lead ECG-AI models showed strong predictive
Networks (CNNs) and Recurrent Neural Networks (RNNs) to performance. The top model, known as C12-ECG-AI-Cox,
effectively extract features from RR intervals. Notably, the performed exceptionally well by integrating demographic
model demonstrates superior performance compared to existing information, clinical data, and ECG-AI, indicating its potential
methodologies, achieving high sensitivity and specificity across for clinical use to identify individuals at high risk. Detailed
three different databases. The preprocessing phase involves subgroup analyses demonstrated that the model's ability to
converting raw ECG data into RR intervals, followed by predict FCHD risk was consistent across various genders,
segmentation. The model architecture includes convolutional races/ethnicities, and patients with existing coronary artery
layers for feature extraction and Long Short-Term Memory disease. This suggests that the model is robust and widely
(LSTM) layers for capturing temporal dependencies, applicable. Additionally, certain age groups showed even higher
complemented by a pooling layer for downsampling. This accuracy in predicting cardiac events, highlighting the
combination allows the model to process 24-hour ECG effectiveness of ECG-based models for specific age
recordings in less than a second, showcasing its computational demographics in risk assessment. The research also emphasizes
efficiency. Overall, the proposed model not only surpasses state- the critical role of ECG-based predictions in clinical practice.
of-the-art methods in AF detection but also ensures rapid and These models can identify high-risk individuals, allowing for
reliable analysis, making it a significant advancement in the field early interventions and customized preventive strategies. The
of cardiovascular health monitoring. This algorithm was tested findings indicate that ECG-AI models could significantly
on the unseen datasets to examine its robustness in detecting AF contribute to personalized medicine, aiding clinicians in making
(Atrial Fibrillation) for new recordings which resulted in 98.96% more informed decisions and improving patient outcomes. In
and 86.04% for specificity and sensitivity, respectively. summary, combining demographics, clinical data, and ECG-AI
into predictive models represents a significant advancement in
7: Machine Learning Algorithms for Cardiovascular assessing cardiovascular risk. The high accuracy and broad
Disease (CVD) Prediction[7] applicability of these models underscore their potential to
enhance preventive healthcare strategies and improve the early
This research article explores the early prediction of detection of fatal coronary heart disease.
cardiovascular disease (CVD) through the application of
machine learning algorithms to health records. It underscores the
significance of incorporating genetic predispositions, lifestyle
choices, and clinical markers into these models to achieve 9: Arrhythmia Classification using Morphological and
precise risk assessments. The study emphasizes how such Dynamic Features[9]
integration can advance preventive healthcare strategies and
facilitate personalized interventions by leveraging electronic This paper introduces a novel approach for arrhythmia
health records (EHRs). Integrating diverse data sources, classification by combining morphological and dynamic features
including genetic information, lifestyle factors, and clinical extracted from ECG signals. Morphological features are derived
markers, into machine learning models enhances the preemptive using Wavelet Transform and Independent Component Analysis,
risk assessment for CVD. Advanced machine learning while dynamic features are obtained from RR interval
technologies play a pivotal role in refining these risk assessments information. For classification, a Support Vector Machine
and improving the predictive accuracy of CVD outcomes. (SVM) is employed, categorizing the data into 15 distinct classes
Critical steps such as data preprocessing and feature selection are and achieving an impressive overall accuracy of 99.66% on a
essential in preparing EHRs for effective machine-learning dataset of 85,945 heartbeats. The method is validated using the
applications. Various machine learning models, including logistic MIT-BIH Arrhythmias Database, demonstrating superior
regression, decision trees, random forests, support vector performance compared to previous methodologies. To further
machines, and neural networks, have demonstrated efficacy in enhance accuracy, the study employs two fusion methods to
predicting CVD. Utilizing EHRs for prediction supports the integrate results from two ECG leads. This combination of
paradigm shift towards precision medicine, enabling morphological and dynamic features, along with the use of
individualized risk assessments and proactive interventions. This fusion techniques, significantly improves the robustness and
approach not only improves the accuracy of CVD predictions but precision of arrhythmia classification. The approach's high
also aligns with modern healthcare's movement towards more accuracy underscores its potential for clinical application in
personalized and preventive care strategies. reliable arrhythmia detection.
 10: Deep Convolutional Neural Networks for Heartbeat xresnet1d101 model demonstrated superior adaptability when
Classification[10] fine-tuned on smaller datasets like ICBEB2018, illustrating the
effectiveness of pretraining on comprehensive datasets for
This paper introduces a method utilizing deep convolutional specific clinical applications [1].
neural networks (CNNs) for classifying heartbeats, specifically
identifying five different types of arrhythmias. Additionally, it Moreover, the emergence of innovative AI architectures like
explores a technique for transferring the knowledge gained from Kolmogorov-Arnold Networks (KANs) presents promising
this task to the classification of myocardial infarction, achieving alternatives to traditional MLPs, offering improved
high accuracy in both classifications. The study focuses on deep interpretability and computational efficiency. While KANs are
learning approaches for precise heartbeat classification and relatively new in ECG feature extraction, their potential for
effective knowledge transfer between different cardiac tasks. The enhancing accuracy and scalability suggests a fertile area for
method demonstrates impressive accuracy in categorizing both future research and development [3]...
arrhythmias and myocardial infarction. Preprocessing steps
Further development in ECG telemonitoring systems, leveraging
include cropping individual beats and applying zero padding to
wearable wireless sensor networks (WSNs) and IoT integration,
ensure uniform length. The deep learning framework allows the
holds promise for advancing personalized healthcare solutions.
model to learn task-specific features, requiring a large dataset for
These systems not only facilitate real-time monitoring but also
effective training. The research utilizes labeled ECG records
empower patients with continuous health data access, reducing
from the PhysioNet MIT-BIH Arrhythmia Database and the PTB
dependency on traditional clinical visits and enhancing proactive
Diagnostic ECG Database, providing a robust foundation for
disease management [4][5].
training and validation. The results highlight the effectiveness of
deep learning in improving the accuracy and reliability of
Looking ahead, future research directions should focus on
cardiac condition classifications.
addressing current challenges such as noise reduction, signal
artifacts, and the integration of multi-modal health data (genetic,
III. CONCLUSION
lifestyle, and clinical) into AI models. Additionally, exploring
In conclusion, this literature review has underscored the explainable AI methodologies will be crucial for gaining insights
significant advancements and diverse applications of artificial into model decisions, thereby fostering trust and adoption in
intelligence (AI) in ECG analysis and cardiovascular health clinical settings[7][10].
monitoring. The studies reviewed highlight the efficacy of deep
In summary, the convergence of AI with ECG analysis is poised
learning models, particularly convolutional neural networks
to revolutionize cardiovascular healthcare by enabling early
(CNNs) and their variants, in various aspects of ECG
detection, personalized risk assessment, and targeted
interpretation, including arrhythmia detection, atrial fibrillation
interventions. Continued interdisciplinary collaboration and
(AF) diagnosis, and risk prediction for cardiovascular diseases
technological innovation will drive further advancements,
(CVD) such as coronary heart disease (FCHD) [[1][6][8][10] ]..
ultimately improving patient outcomes and healthcare efficiency
The integration of large-scale datasets like PTB-XL and MIT- on a global scale.
BIH has enabled the development of robust models through
techniques such as transfer learning, enhancing the accuracy and
generalization of ECG-based AI systems. For instance, the

Reference

[1] N. Strodthoff, P. Wagner, T. Schaeffter, and W. Samek, “Deep Z. Liu et al., “KAN: Kolmogorov-Arnold Networks,” Apr. 2024,
Learning for ECG Analysis: Benchmarks and Insights from [Online]. Available: http://arxiv.org/abs/2404.19756
PTB-XL,” IEEE J Biomed Health Inform, vol. 25, no. 5, pp.
1519–1528, May 2021, doi: 10.1109/JBHI.2020.3022989. A. El Attaoui, M. Hazmi, A. Jilbab, and A. Bourouhou, “Wearable
Wireless Sensors Network for ECG Telemonitoring Using Neural
Network for Features Extraction,” Wirel Pers Commun, vol. 111, no. 3,
[2] M. A. Serhani, H. T. El Kassabi, H. Ismail, and A. N. Navaz, “ECG
pp. 1955–1976, Apr. 2020, doi: 10.1007/s11277-019-06967-x.
monitoring systems: Review, architecture, processes, and key
challenges,” Sensors (Switzerland), vol. 20, no. 6. MDPI AG, Mar. 02,
A. P. Jawalkar et al., “Early prediction of heart disease with data analysis
2020. doi: 10.3390/s20061796.
using supervised learning with stochastic gradient boosting,” Journal of
Engineering and Applied Science, vol. 70, no. 1, Dec. 2023, doi:
10.1186/s44147-023-00280-y.
[6] R. S. Andersen, A. Peimankar, and S. Puthusserypady, “A deep learning
approach for real-time detection of atrial fibrillation,” Expert Syst Appl,
vol. 115, pp. 465–473, Jan. 2019, doi: 10.1016/j.eswa.2018.08.011. IEEE Staff and IEEE Staff, 2010 32nd Annual International Conference
of the IEEE Engineering in Medicine and Biology Society.
[7] D. R. Deepa, V. B. Sadu, C. G. Prashant, and D. A. Sivasamy, “Early
prediction of cardiovascular disease using machine learning: Unveiling [10] M. U. Sreeja and M. H. Supriya, “A Deep Convolutional Model for
risk factors from health records,” AIP Adv, vol. 14, no. 3, Mar. 2024, doi: Heart Disease Prediction based on ECG Data with Explainable AI,”
10.1063/5.0191990. WSEAS Transactions on Information Science and Applications, vol. 20,
pp. 254–264, 2023, doi: 10.37394/23209.2023.20.29.
[8] O. Akbilgic, “Time-Dependent ECG-AI Prediction of Fatal Coronary
Heart Disease”, doi: 10.1101/2023.10.11.23296910.