What question did this study set out to answer?

To develop and evaluate a lightweight two-dimensional convolutional neural network capable of real-time multi-class ECG arrhythmia classification on edge-AI healthcare devices.

August 13, 2026Systems and Soft ComputingOpen Access

Lightweight 2D CNN achieves ~86% accuracy for ECG arrhythmia classification with low computational demand.

Q: What is the clinical evidence from this study?

Study design: Other. Population: ECG arrhythmia. Intervention: Lightweight 2D CNN using high-resolution STFT spectrograms vs. Representative deep CNN architectures and lightweight models. Primary outcome: Accuracy.

View Full Paper

Ask AI

Bookmark

Why the study?

Accurate and lightweight ECG arrhythmia classification frameworks suitable for real-time deployment in wearable devices, embedded systems, and edge-AI healthcare applications were needed.

Population

ECG segments containing five consecutive R-peaks

Comparison

Proposed lightweight 2D CNN vs representative deep and lightweight CNN architectures

Design

Algorithm development and validation study

Key result

A proposed lightweight 2D CNN using high-resolution STFT spectrograms achieved an accuracy of 86.16% and specificity of 96.54% for ECG arrhythmia classification with low computational requirements.

Authors

CLChun‐Ling Lin MSMing-Chih Shen MLMeng-Shen Li

Discussion

Member takes

Overview

May facilitate efficient arrhythmia detection in low-resource settings; leaves open prospective clinical validation before adoption.

Key Points

To develop and evaluate a lightweight two-dimensional convolutional neural network capable of real-time multi-class ECG arrhythmia classification on edge-AI healthcare devices.
Constructed a six-block 2D CNN trained on high-resolution short-time Fourier transform (STFT) spectrograms generated from ECG segments with five consecutive R-peaks.
Evaluated performance using standard train–validation–test partitioning and record-wise three-fold cross-validation against standard deep and lightweight networks, including VGG, ResNet, MobileNetV2, and ShuffleNetV2.
Under the train–validation–test split, the proposed architecture achieved an accuracy of 86.16%, an F1-score of 86.07%, and a specificity of 96.54% with 898,413 trainable parameters.
Under record-wise three-fold cross-validation, the model attained the highest macro AUROC among evaluated lightweight networks while requiring 0.71 GFLOPs.
Ablation testing demonstrated that high-resolution STFT representations and five-consecutive-R-peak segmentations substantially outperformed raw one-dimensional ECG inputs and single-beat models.

Structured PICO

Population

ECG segments containing five consecutive R-peaks

Intervention

Lightweight two-dimensional convolutional neural network (2D CNN) using high-resolution short-time Fourier transform (STFT) spectrograms

Comparator

Representative deep CNN architectures and lightweight models (VGG16, VGG19, ResNet variants, Inception-ResNet V2, DenseNet, Autoencoder-based networks, MobileNetV2, EfficientNet-B0, and ShuffleNetV2)

Outcome

Classification accuracy, F1-score, specificity, and macro AUROCsurrogate

A proposed lightweight 2D CNN using STFT spectrograms achieves high accuracy for ECG arrhythmia classification while maintaining low computational requirements suitable for wearable devices.

Cite This Study

Lin et al. (2026) studied ECG arrhythmia. Lightweight 2D CNN using high-resolution STFT spectrograms vs. Representative deep CNN architectures and lightweight models was evaluated on Accuracy. A proposed lightweight 2D CNN using high-resolution STFT spectrograms achieved an accuracy of 86.16% and specificity of 96.54% for ECG arrhythmia classification with low computational requirements.

synapsesocial.com/papers/6a9d7d829ce529ba6dfce86a https://doi.org/10.1016/j.sasc.2026.200598

View Full Paper

Ask AI

Bookmark

August 13, 2026Systems and Soft ComputingOpen Access

Lightweight 2D CNN achieves ~86% accuracy for ECG arrhythmia classification with low computational demand.

View Full Paper

Ask AI

Bookmark

Why the study?

Accurate and lightweight ECG arrhythmia classification frameworks suitable for real-time deployment in wearable devices, embedded systems, and edge-AI healthcare applications were needed.

Population

ECG segments containing five consecutive R-peaks

Comparison

Proposed lightweight 2D CNN vs representative deep and lightweight CNN architectures

Design

Algorithm development and validation study

Key result

A proposed lightweight 2D CNN using high-resolution STFT spectrograms achieved an accuracy of 86.16% and specificity of 96.54% for ECG arrhythmia classification with low computational requirements.

Authors

CLChun‐Ling Lin MSMing-Chih Shen MLMeng-Shen Li

Discussion

Member takes

Overview

May facilitate efficient arrhythmia detection in low-resource settings; leaves open prospective clinical validation before adoption.

Key Points

To develop and evaluate a lightweight two-dimensional convolutional neural network capable of real-time multi-class ECG arrhythmia classification on edge-AI healthcare devices.
Constructed a six-block 2D CNN trained on high-resolution short-time Fourier transform (STFT) spectrograms generated from ECG segments with five consecutive R-peaks.
Evaluated performance using standard train–validation–test partitioning and record-wise three-fold cross-validation against standard deep and lightweight networks, including VGG, ResNet, MobileNetV2, and ShuffleNetV2.
Under the train–validation–test split, the proposed architecture achieved an accuracy of 86.16%, an F1-score of 86.07%, and a specificity of 96.54% with 898,413 trainable parameters.
Under record-wise three-fold cross-validation, the model attained the highest macro AUROC among evaluated lightweight networks while requiring 0.71 GFLOPs.
Ablation testing demonstrated that high-resolution STFT representations and five-consecutive-R-peak segmentations substantially outperformed raw one-dimensional ECG inputs and single-beat models.

Structured PICO

Population

ECG segments containing five consecutive R-peaks

Intervention

Lightweight two-dimensional convolutional neural network (2D CNN) using high-resolution short-time Fourier transform (STFT) spectrograms

Comparator

Representative deep CNN architectures and lightweight models (VGG16, VGG19, ResNet variants, Inception-ResNet V2, DenseNet, Autoencoder-based networks, MobileNetV2, EfficientNet-B0, and ShuffleNetV2)

Outcome

Classification accuracy, F1-score, specificity, and macro AUROCsurrogate

A proposed lightweight 2D CNN using STFT spectrograms achieves high accuracy for ECG arrhythmia classification while maintaining low computational requirements suitable for wearable devices.

Cite This Study

synapsesocial.com/papers/6a9d7d829ce529ba6dfce86a https://doi.org/10.1016/j.sasc.2026.200598

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

View Full Paper

Ask AI

Bookmark