A Novel Hybrid Deep Learning Framework with Learnable Feature Kernels for Arrhythmia Classification

Abstract Arrhythmia, a common cardiovascular disease, can lead to severe outcomes such as cardiac arrest and stroke. The electrocardiogram (ECG) is an inexpensive and non-invasive test that provides valuable diagnostic information. However, manual analysis of ECG signals is labor-intensive and time-consuming, making it error-prone. Automated ECG-based arrhythmia classification using artificial intelligence (AI) is crucial for early detection of cardiovascular diseases, yet existing models struggle to balance simplicity and effectiveness. This study proposes a hybrid approach that leverages the advantages of both traditional and deep learning methods to improve arrhythmia classification. In the proposed approach, features are developed in two stages: first, classical features are extracted, and then these features, along with ECG heartbeats from two leads, are fed into a convolutional neural network to learn more discriminative representations. Additionally, to partially mitigate class imbalance, a weighted cross-entropy loss function is utilized. The proposed hybrid model was evaluated using two widely used benchmark databases, the MIT-BIH Arrhythmia Database and the St. Petersburg INCART 12-lead Arrhythmia Database, under four evaluation paradigms: intra-subject evaluation, inter-subject evaluation, bidirectional cross-dataset evaluation, and subject-wise k-fold cross-validation. The proposed hybrid model achieved overall accuracies of 99.587% and 91.43% under intra-subject and inter-subject evaluation, respectively, and 94.116% and 86.28% in the MIT-BIH-to-INCART and INCART-to-MIT-BIH cross-dataset directions, respectively. Under subject-wise k-fold cross-validation, the model achieved a mean accuracy of 94.02%. The model also maintained substantially lower computational complexity than the compared deep learning methods. These results suggest that the proposed hybrid model offers a computationally efficient approach to ECG arrhythmia classification, although further improvements in minority-class recognition is needed.

Authors

Institutions

Publication Details

Journal
Engineering Research Express
Published
2026-11-17
DOI
https://doi.org/10.1088/2631-8695/aec36e
Primary Topic
ECG Monitoring and Analysis
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Novel Hybrid Deep Learning Framework with Learnable Feature Kernels for Arrhythmia Classification

Mohammad Norizadeh Cherloo, Mohammad Reza Daliri
4 links
Engineering Research Express
ECG Monitoring and Analysis
article

A Novel Hybrid Deep Learning Framework with Learnable Feature Kernels for Arrhythmia Classification

Mohammad Norizadeh Cherloo, Mohammad Reza Daliri
article en

Abstract

Abstract Arrhythmia, a common cardiovascular disease, can lead to severe outcomes such as cardiac arrest and stroke. The electrocardiogram (ECG) is an inexpensive and non-invasive test that provides valuable diagnostic information. However, manual analysis of ECG signals is labor-intensive and time-consuming, making it error-prone. Automated ECG-based arrhythmia classification using artificial intelligence (AI) is crucial for early detection of cardiovascular diseases, yet existing models struggle to balance simplicity and effectiveness. This study proposes a hybrid approach that leverages the advantages of both traditional and deep learning methods to improve arrhythmia classification. In the proposed approach, features are developed in two stages: first, classical features are extracted, and then these features, along with ECG heartbeats from two leads, are fed into a convolutional neural network to learn more discriminative representations. Additionally, to partially mitigate class imbalance, a weighted cross-entropy loss function is utilized. The proposed hybrid model was evaluated using two widely used benchmark databases, the MIT-BIH Arrhythmia Database and the St. Petersburg INCART 12-lead Arrhythmia Database, under four evaluation paradigms: intra-subject evaluation, inter-subject evaluation, bidirectional cross-dataset evaluation, and subject-wise k-fold cross-validation. The proposed hybrid model achieved overall accuracies of 99.587% and 91.43% under intra-subject and inter-subject evaluation, respectively, and 94.116% and 86.28% in the MIT-BIH-to-INCART and INCART-to-MIT-BIH cross-dataset directions, respectively. Under subject-wise k-fold cross-validation, the model achieved a mean accuracy of 94.02%. The model also maintained substantially lower computational complexity than the compared deep learning methods. These results suggest that the proposed hybrid model offers a computationally efficient approach to ECG arrhythmia classification, although further improvements in minority-class recognition is needed.

Engineering Research Express
Iran University of Science and Technology (IR)
Good health and well-being
ECG Monitoring and Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon