ReHeartNet: Reconstruct Electrocardiogram From Photoplethysmography by Using Dense Connected Deep Learning Model.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Current electrocardiogram (ECG) monitoring requires multiple electrodes, causing discomfort and skin irritation.
• Photoplethysmogram (PPG) signals offer a non-invasive alternative for physiological monitoring.
• Accurate reconstruction of ECG from PPG is crucial for convenient cardiac monitoring.

Purpose of the Study:

• To develop a novel neural network, ReHeartNet, for high-fidelity ECG signal reconstruction from PPG signals.
• To eliminate the need for invasive ECG electrodes, improving patient comfort and compliance.
• To enable non-invasive, continuous cardiac rhythm monitoring using wearable PPG sensors.

Main Methods:

• ReHeartNet formulates ECG reconstruction from PPG as a regression problem.
• The model utilizes densely connected bidirectional long short-term memory (DC-BiLSTM) blocks to capture multi-scale temporal and frequency relationships.
• Hierarchical features from different BiLSTM layers are fused to enhance reconstruction accuracy.

Main Results:

• ReHeartNet demonstrated superior performance in ECG reconstruction across four diverse datasets (MIMIC-III, BIDMC, TBME-RR, CBIC-Heart).
• The proposed method consistently outperformed existing baseline models, including Generative Adversarial Networks (GAN), Recurrent Neural Networks (RNN), and Transformers.
• Experiments validated the model's effectiveness in reconstructing ECG signals from PPG data.

Conclusions:

• ReHeartNet exhibits strong generalization and robustness for ECG reconstruction using only wearable PPG signals.
• The method is effective for both healthy individuals and patients with circulatory diseases and arrhythmias.
• This technology supports reliable cardiac monitoring across various populations, enhancing non-invasive healthcare solutions.