Dual-Path Cuffless PPG-Based Blood Pressure Estimation Using Conformer & Swin Transformer | JoVE Visualize

Area of Science:

Biomedical Engineering
Artificial Intelligence
Cardiovascular Technology

Background:

Continuous, non-invasive blood pressure (BP) monitoring remains a significant challenge in healthcare.
Existing methods often rely on invasive procedures or less accurate estimations.
Photoplethysmogram (PPG) signals offer a promising avenue for cuffless BP monitoring due to their accessibility.

Purpose of the Study:

To introduce a novel dual-path deep learning framework for continuous, non-invasive cuffless blood pressure monitoring.
To adapt and apply two novel deep neural network architectures, Conformer-Transformer and 1D Swin Transformer, to PPG signal analysis.
To enhance the accuracy and physiological consistency of Systolic Blood Pressure (SBP) and Diastolic Blood Pressure (DBP) estimation.

Main Methods:

Development of a dual-path deep learning framework utilizing PPG signals.
Implementation and adaptation of Conformer-Transformer and 1D Swin Transformer architectures.
Modeling of both morphological structure and rhythmic dynamics of PPG signals for Arterial Blood Pressure (ABP) waveform reconstruction.
Extensive experimental validation on two public datasets.

Main Results:

The Conformer-Transformer achieved the lowest Mean Absolute Error (MAE) for SBP (2.979 mmHg) and DBP (1.603 mmHg), outperforming previous studies.
The 1D Swin Transformer demonstrated strong performance with a systolic MAE of 3.034 mmHg and diastolic MAE of 1.714 mmHg.
Both models significantly improved BP estimation accuracy and physiological consistency, meeting BHS grade A and AAMI standards.
Superior Arterial Blood Pressure (ABP) waveform reconstruction was achieved.

Conclusions:

The proposed dual-path deep learning framework, incorporating Conformer-Transformer and 1D Swin Transformer, offers a significant advancement in cuffless blood pressure monitoring.
These novel architectures effectively model PPG signal characteristics for accurate and reliable SBP and DBP estimation.
The study demonstrates the potential of advanced deep learning techniques for non-invasive cardiovascular monitoring, aligning with stringent clinical standards.