Arterial blood pressure waveform reconstruction estimation from PPG using TCN-BiLSTM

Area of Science:

• Biomedical Engineering
• Cardiovascular Physiology
• Artificial Intelligence in Medicine

Background:

• Photoplethysmography (PPG) offers non-invasive, continuous blood pressure (BP) monitoring.
• Challenges include noise, motion artifacts, and inter-individual variability affecting PPG accuracy.
• Accurate reconstruction of arterial blood pressure (ABP) waves from PPG is crucial for reliable BP estimation.

Purpose of the Study:

• To develop and validate a deep learning framework for reconstructing ABP waveforms from PPG signals.
• To enable accurate estimation of systolic and diastolic blood pressure (SBP/DBP) without a cuff.
• To assess the model's performance on independent datasets for robust validation.

Main Methods:

• Implementation of a Temporal Convolutional Network (TCN) and Bidirectional Long Short-Term Memory (BiLSTM) model (TCN-BiLSTM).
• Utilizing derivative-enhanced PPG representations to capture cardiovascular dynamics.
• Training and validation on the VitalDB database (2226 patients) and external testing on the MIMIC-IV database.

Main Results:

• Achieved Mean Absolute Error (MAE) of 3.61 mmHg for SBP and 1.79 mmHg for DBP.
• Achieved Root Mean Square Error (RMSE) of 5.24 mmHg for SBP and 2.37 mmHg for DBP.
• Demonstrated consistent performance on the independent MIMIC-IV cohort without retraining.

Conclusions:

• The TCN-BiLSTM deep learning framework accurately reconstructs ABP waveforms from PPG.
• This approach provides a physiologically valid and accurate method for continuous, cuffless blood pressure estimation.
• The study confirms the feasibility of PPG-based ABP reconstruction for practical clinical applications.