Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PPG-to-ECG Signal Translation for Continuous Atrial Fibrillation Detection via Attention-based Deep State-Space Modeling

Khuong Vo, Mostafa El-Khamy, Yoojin Choi

TL;DR

This work tackles translating noninvasive PPG signals into ECG waveforms to enable continuous AFib monitoring. It introduces an attention-based deep state-space model (ADSSM) with a probabilistic prior that learns latent ECG dynamics from PPG in a subject-independent manner, using an attention mechanism to align PP intervals with RR outputs. The model achieves high-fidelity ECG translation (ρ ≈ 0.86, RMSE ≈ 0.07, SNR ≈ 15 dB) and enables AFib detection with nearly real-ECG performance (PR-AUC ≈ 0.986, ROC-AUC ≈ 0.99), even under noise and partial data scenarios. These results suggest practical potential for wearable-based, real-time cardiovascular screening, combining the rich knowledge base of ECG with the continuous measurement capabilities of PPG, and point to future work on uncertainty quantification and broader signal translations.

Abstract

Photoplethysmography (PPG) is a cost-effective and non-invasive technique that utilizes optical methods to measure cardiac physiology. PPG has become increasingly popular in health monitoring and is used in various commercial and clinical wearable devices. Compared to electrocardiography (ECG), PPG does not provide substantial clinical diagnostic value, despite the strong correlation between the two. Here, we propose a subject-independent attention-based deep state-space model (ADSSM) to translate PPG signals to corresponding ECG waveforms. The model is not only robust to noise but also data-efficient by incorporating probabilistic prior knowledge. To evaluate our approach, 55 subjects' data from the MIMIC-III database were used in their original form, and then modified with noise, mimicking real-world scenarios. Our approach was proven effective as evidenced by the PR-AUC of 0.986 achieved when inputting the translated ECG signals into an existing atrial fibrillation (AFib) detector. ADSSM enables the integration of ECG's extensive knowledge base and PPG's continuous measurement for early diagnosis of cardiovascular disease.

PPG-to-ECG Signal Translation for Continuous Atrial Fibrillation Detection via Attention-based Deep State-Space Modeling

TL;DR

This work tackles translating noninvasive PPG signals into ECG waveforms to enable continuous AFib monitoring. It introduces an attention-based deep state-space model (ADSSM) with a probabilistic prior that learns latent ECG dynamics from PPG in a subject-independent manner, using an attention mechanism to align PP intervals with RR outputs. The model achieves high-fidelity ECG translation (ρ ≈ 0.86, RMSE ≈ 0.07, SNR ≈ 15 dB) and enables AFib detection with nearly real-ECG performance (PR-AUC ≈ 0.986, ROC-AUC ≈ 0.99), even under noise and partial data scenarios. These results suggest practical potential for wearable-based, real-time cardiovascular screening, combining the rich knowledge base of ECG with the continuous measurement capabilities of PPG, and point to future work on uncertainty quantification and broader signal translations.

Abstract

Photoplethysmography (PPG) is a cost-effective and non-invasive technique that utilizes optical methods to measure cardiac physiology. PPG has become increasingly popular in health monitoring and is used in various commercial and clinical wearable devices. Compared to electrocardiography (ECG), PPG does not provide substantial clinical diagnostic value, despite the strong correlation between the two. Here, we propose a subject-independent attention-based deep state-space model (ADSSM) to translate PPG signals to corresponding ECG waveforms. The model is not only robust to noise but also data-efficient by incorporating probabilistic prior knowledge. To evaluate our approach, 55 subjects' data from the MIMIC-III database were used in their original form, and then modified with noise, mimicking real-world scenarios. Our approach was proven effective as evidenced by the PR-AUC of 0.986 achieved when inputting the translated ECG signals into an existing atrial fibrillation (AFib) detector. ADSSM enables the integration of ECG's extensive knowledge base and PPG's continuous measurement for early diagnosis of cardiovascular disease.
Paper Structure (16 sections, 14 equations, 4 figures, 2 tables)

This paper contains 16 sections, 14 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. State-Space Modeling of ECG from PPG Signals
  4. ECG Generative (Decoding) Process from PPG
  5. Latent State Inference (Posterior Encoding) Process
  6. Training Process
  7. Neural Network Architectures
  8. Experiments
  9. Dataset
  10. Evaluation Metrics
  11. ECG Translation from PPG
  12. AFib Detection
  13. Implementation and Results
  14. ECG Translation from PPG
  15. AFib Detection
  16. ...and 1 more sections

Figures (4)

  • Figure 1: A PPG-ECG waveform pair. PPG signals can often become contaminated by noise.
  • Figure 2: The graphical model for ECG translation from PPG. Shaded nodes represent observed variables. Clear nodes represent latent variables. Diamond nodes denote deterministic variables. Variables $\mathbf{x}_t, \mathbf{y}_t$, and $\mathbf{c}_t$ represent PP intervals, RR intervals, and context vectors, respectively. $\alpha_{t,i}$ are attention weights defines how well two intervals $\mathbf{x}_i$ and $\mathbf{y}_t$ are aligned. The attention mechanism is shown only at time step 2.
  • Figure 3: The graphical model at latent state inference time. Variables $\mathbf{y}_t, \mathbf{h}_t, \mathbf{g}_t$, and $\mathbf{z}_t$ represent respectively RR intervals, backward, forward recurrent states, and latent states.
  • Figure 4: Examples of the translated ECG signals. In each subfigure: the top panel shows the input PPG waveform and the bottom panel shows the reconstructed ECG waveform compared with the reference waveform. The average ECG waveform (dark blue) of all possible pulses overlaid on each individual pulse (light blue).