Cardiac-Net: A Wearable ECG Prototype and Hybrid CNN–Transformer Model for Atrial Fibrillation Detection in Long-Term ECG Recordings

Q: What is the clinical evidence from this study?

Population: Atrial Fibrillation (n=84). Intervention: Cardiac-Net wearable ECG system and hybrid 1D CNN-Transformer classifier vs. Standard BIOPAC MP36 reference (bench testing); independent MIT-BIH database (zero-shot evaluation). Primary outcome: Atrial fibrillation detection accuracy (AUC).

Q: What does this research mean for the field?

A wearable ECG acquisition system paired with a hybrid CNN–Transformer model achieves high segment-level discrimination for atrial fibrillation (AUC = 0.953) but exhibits substantial false-alarm burden and patient-level performance variation during continuous monitoring. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Aymane Edder; Fatima-Ezzahraa Ben-Bouazza; Youssef Ait Bigane; Djeneba Sangare; Idriss Tafala; Oumaima Manchadi; Bassma Jioudi

doi:10.3390/biomedinformatics6050090

October 12, 2026BioMedInformaticsOpen Access

Cardiac-Net: A Wearable ECG Prototype and Hybrid CNN–Transformer Model for Atrial Fibrillation Detection in Long-Term ECG Recordings

View Full Paper

Ask AI

Bookmark

Share

Key result

Cardiac-Net detects AF with ~0.95 AUC, but high false-alarm rates hinder clinical deployment.

n=84

Why the study?

Postoperative atrial fibrillation is a leading cause of stroke and mortality after cardiac surgery, motivating continuous long-term ECG monitoring.

Does the Cardiac-Net wearable ECG system and hybrid CNN-Transformer model accurately detect atrial fibrillation in long-term ECG recordings?

Key Points

Develop and validate Cardiac-Net, a wearable ECG hardware prototype integrated with a hybrid 1D CNN–Transformer model for detecting postoperative atrial fibrillation during continuous long-term monitoring.
Bench-tested an ESP32-S3/AD8232 BLE sensor prototype against a BIOPAC MP36 reference across five simulated noise conditions using a Fluke ProSim 8 simulator.
Trained and evaluated a 1D CNN–Transformer classifier on the Long-Term Atrial Fibrillation Database (N=84 recordings) using strict 5-fold patient-wise cross-validation with zero recording overlap.
Conducted zero-shot domain evaluation on the independent MIT-BIH Atrial Fibrillation Database without fine-tuning, alongside Grad-CAM and Integrated Gradients model interpretability analyses.
Hardware bench testing achieved a Pearson Correlation Coefficient of 0.9222 and heart rate errors below 0.2% under clean conditions, meeting IEC 60601-2-47 standards.
Five-fold cross-validation demonstrated high discrimination with AUC = 0.953±0.016, F1 = 0.894±0.030, precision = 0.880±0.051, and recall = 0.915±0.073 (AUC = 0.946±0.010 at 85 Hz BLE throughput), alongside a zero-shot ensemble AUC of 0.834 on MIT-BIH.
Episode-level sensitivity reached a per-patient median of 1.000 (IQR [0.925, 1.000]), but 7.2% of patients fell below chance detection while generating 86±38 false alarms per patient-day, reducing positive predictive value to 0.28–0.44 at 5–10% assumed prevalence.

Structured PICO

Does the Cardiac-Net wearable ECG system and hybrid CNN-Transformer model accurately detect atrial fibrillation in long-term ECG recordings?

P

Population

84 long-term ECG recordings utilized to train and validate a wearable ECG prototype and hybrid CNN-Transformer model for atrial fibrillation detection.

E

Exposure

Cardiac-Net, a wearable ECG system (ESP32-S3/AD8232 BLE device) combined with a hybrid 1D CNN-Transformer classifier.

O

Outcome

Atrial fibrillation detection performance measured by Area Under the Curve (AUC), F1 score, precision, and recall.

Limitations

The cloud inference and monitoring-dashboard layer is a design-level proposal only and has not been implemented.
Detection accuracy is not uniform across patients, with 7.2% falling below chance-level detection.
High false-alarm burden of 86±38 alerts per patient-day.
Prevalence-adjusted positive predictive value is only 0.28–0.44 at 5–10% assumed AF prevalence.
Alarm calibration and patient-level reliability remain barriers to unsupervised clinical deployment.
Measurable domain gap when generalizing to an independent dataset.
High false-alarm burden (86±38 alerts per patient-day)
Low prevalence-adjusted positive predictive value (0.28-0.44 at 5-10% assumed AF prevalence)
7.2% of patients fall below chance-level detection
Alarm calibration and patient-level reliability remain barriers to unsupervised clinical deployment

Implication

The Cardiac-Net prototype achieves high segment-level diagnostic accuracy for atrial fibrillation detection, but high false-alarm rates must be addressed prior to unsupervised clinical deployment.

Abstract

Postoperative atrial fibrillation (POAF) is a leading cause of stroke and mortality after cardiac surgery, motivating continuous long-term ECG monitoring.This paper presents Cardiac-Net, a wearable ECG system (ESP32-S3/AD8232 BLE device with a Flutter mobile gateway; 128 Hz nominal/85 Hz effective BLE sampling rate, 12.6 ms mean latency, zero over-the-air loss of transmitted packets, 3.11 mJ per transmitted sample) combined with a hybrid 1D CNN–Transformer classifier and a proposed cloud-assisted monitoring architecture; the cloud inference and monitoring-dashboard layer is a design-level proposal only and has not been implemented. Bench signal fidelity, validated against a BIOPAC MP36 reference using a Fluke ProSim 8 simulator across five noise conditions, yields a Pearson Correlation Coefficient of 0.9222 and BPM errors below 0.2% under clean conditions, within the 5% heart-rate-error threshold specified by IEC 60601-2-47. The classifier was trained and evaluated on the Long-Term Atrial Fibrillation Database (84 recordings) using a record-wise (patient-wise) data split with verified zero overlap between training, validation, and test recordings; 5-fold record-wise cross-validation spanning all 84 recordings yields AUC =0.953±0.016, F1 =0.894±0.030, precision =0.880±0.051, and recall =0.915±0.073; re-evaluating at the device’s actual 85 Hz effective throughput confirms comparable discrimination (AUC =0.946±0.010) alongside a real increase in false-alarm burden. A clinical evaluation protocol beyond segment-level accuracy shows that most AF episodes are detected (episode-level sensitivity 0.839±0.222 by fold; per-patient median 1.000, IQR [0.925, 1.000]), but this is not uniform across patients: 7.2% (6/83) fall below chance-level detection. Combined with a false-alarm burden of 86±38 alerts per patient-day and a prevalence-adjusted positive predictive value of only 0.28–0.44 at 5–10% assumed AF prevalence, alarm calibration and patient-level reliability both remain barriers to unsupervised clinical deployment. Zero-shot evaluation on the independent MIT-BIH Atrial Fibrillation Database, with no fine-tuning, yields an ensemble AUC of 0.834, indicating generalization beyond the training distribution alongside a measurable domain gap. Post-hoc Grad-CAM and Integrated Gradients attributions are consistent with physiologically grounded decision-making, concentrating on RR-interval irregularity and QRS morphology. Cardiac-Net thus contributes a validated wearable acquisition layer and a rigorously, record-wise cross-validated detection model, together with an honest characterization of the false-alarm and calibration challenges that remain before postoperative clinical deployment.

Discussion

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

Cite This Study

Edder et al. (2026) studied Atrial Fibrillation (n=84). Cardiac-Net wearable ECG system and hybrid 1D CNN-Transformer classifier vs. Standard BIOPAC MP36 reference (bench testing); independent MIT-BIH database (zero-shot evaluation) was evaluated on Atrial fibrillation detection accuracy (AUC). The Cardiac-Net hybrid CNN-Transformer classifier achieved an AUC of 0.953 for atrial fibrillation detection in cross-validation, but high false-alarm burdens remain a barrier to deployment.

synapsesocial.com/papers/6ace6e0db52af824ab02bcb0 https://doi.org/10.3390/biomedinformatics6050090