Automated Atrial Fibrillation Detection with ECG

Q: What is the clinical evidence from this study?

Population: Atrial fibrillation (n=5925). Intervention: Fine-tuned EfficientNet B0 deep-learning model using ECG spectrograms vs. Ground truth ECG classification (atrial fibrillation vs normal). Primary outcome: Binary classification performance for atrial fibrillation detection on the test set (N=592).

Q: What does this research mean for the field?

A fine-tuned EfficientNet B0 deep learning model trained on ECG spectrogram representations reliably detects atrial fibrillation, achieving up to 97.3% accuracy and an 88.2% F1 score under class imbalance. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Ting-Ruen Wei; Senbao Lu; Yuling Yan

doi:10.3390/bioengineering9100523

October 5, 2022BioengineeringOpen Access

Automated Atrial Fibrillation Detection with ECG

View Full Paper

Ask AI

Bookmark

Share

Key result

ECG spectrogram deep learning detects atrial fibrillation with ~97% accuracy.

n=5,925

Figures

Why the study?

Does a fine-tuned EfficientNet B0 deep-learning algorithm accurately detect atrial fibrillation from ECG signals?

Key Points

To develop a fast, automated deep-learning classification pipeline using spectrogram representations and a fine-tuned convolutional neural network to detect atrial fibrillation from electrocardiography signals.
Converted pre-processed raw electrocardiography (ECG) signals into spectrogram visual representations to mitigate signal noise.
Fine-tuned EfficientNet B0, a pre-trained convolutional neural network, using transfer learning across various dropout rates and weight-initialization strategies.
Evaluated classifier performance on a test set of 592 samples using accuracy, F1 score, Matthews correlation coefficient (MCC), receiver operating characteristic curves, and precision-recall curves.
The model achieved an F-1 score of 88.2% and an accuracy of 97.3% as reported in the primary algorithm presentation.
In test-set evaluation (N=592), the classifier recorded 59 true positives, 11 false positives, 514 true negatives, and 8 false negatives, yielding an accuracy of 96.79%, precision of 84.29%, recall of 88.06%, F1 score of 86.13%, and an MCC of 84.34%.
Model evaluation produced an AUC of 95.34% and an AUC-PR of 92.21%, with transfer learning and fine-tuning showing greater optimization stability than random weight initialization, and a dropout rate of 0.8 producing optimal performance.

Structured PICO

Does a fine-tuned EfficientNet B0 deep-learning algorithm accurately detect atrial fibrillation from ECG signals?

P

Population

5,925 ECG signal samples (5,154 normal and 771 atrial fibrillation) of 9 to 61 seconds duration from the PhysioNet Computing in Cardiology Challenge 2017 dataset, split into 4,534 training, 799 validation, and 592 testing samples.

I

Intervention

Automated deep-learning algorithm (fine-tuned EfficientNet B0) using spectrogram representation of ECG signals

O

Outcome

Prediction of atrial fibrillation (F-1 score and accuracy)

A fine-tuned EfficientNet B0 deep learning model using spectrogram representations of ECG signals can accurately and efficiently detect atrial fibrillation.

Limitations

Mild dataset imbalance with normal samples representing 87% of the dataset (5,154 normal vs 771 atrial fibrillation samples)
Designed only as a binary classifier between atrial fibrillation and normal ECGs, requiring retraining and validation for multi-class or other arrhythmia classifications
Sample size limited to 5,925 ECG recordings; increasing precision curves during fine-tuning suggest a larger dataset is needed to further improve performance
Susceptible to false positives from transient rhythm irregularities in normal ECGs and false negatives in atrial fibrillation ECGs with relatively consistent rhythm

Ask about this paper

Answers are grounded in this paper and cite their sources.

Implication

May enable scalable AF screening; extends ECG AI methods but leaves open prospective clinical validation.

Abstract

An electrocardiography system records electrical activities of the heart, and it is used to assist doctors in the diagnosis of cardiac arrhythmia such as atrial fibrillation. This study presents a fast, automated deep-learning algorithm that predicts atrial fibrillation with excellent performance (F-1 score 88.2% and accuracy 97.3%). Our approach involves the pre-processing of ECG signals, followed by an alternative representation of the signals using a spectrogram, which is then fed to a fine-tuned EfficientNet B0, a pre-trained convolution neural network model, for the classification task. Using the transfer learning approach and with fine-tuning of the EfficientNet, we optimize the model to achieve highly efficient and effective classification of the atrial fibrillation.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Wei et al. (2022) studied Atrial fibrillation (n=5,925). Fine-tuned EfficientNet B0 deep-learning model using ECG spectrograms vs. Ground truth ECG classification (atrial fibrillation vs normal) was evaluated on Binary classification performance for atrial fibrillation detection on the test set (N=592). A fine-tuned EfficientNet B0 model using ECG spectrograms detected atrial fibrillation with up to 97.3% accuracy and an 88.2% F1 score (96.79% accuracy, 86.13% F1, and 84.34% MCC in baseline testing).

synapsesocial.com/papers/6a778a30e29fd788c9f76078 https://doi.org/10.3390/bioengineering9100523

View Full Paper

Ask AI

Bookmark

Share

Automated Atrial Fibrillation Detection with ECG

Figures

Key Points

Structured PICO

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Automated Atrial Fibrillation Detection with ECG

Figures

Key Points

Structured PICO

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading