BM-BPW: A Bidirectional Mamba-based Model for Blood Pressure Waveform Estimation in Arrhythmia

Area of Science:

• Biomedical Engineering
• Cardiovascular Technology
• Artificial Intelligence in Healthcare

Background:

• Continuous blood pressure (BP) monitoring is vital for managing patients with arrhythmia.
• Cuffless BP measurement using electrocardiogram (ECG) and photoplethysmogram (PPG) signals with deep learning shows promise.
• Existing deep learning models struggle with position-sensitive physiological data and global context, limiting accuracy.

Purpose of the Study:

• To introduce BM-BPW, a novel deep learning model for continuous BP waveform estimation from ECG and PPG signals.
• To address the challenges of position-sensitivity and global context in BP monitoring for arrhythmia patients.
• To improve the accuracy and efficiency of cuffless BP monitoring in this population.

Main Methods:

• Developed the BM-BPW model utilizing a bidirectional Mamba architecture with position embeddings for global context and local awareness.
• Integrated a multi-scale convolutional neural network (CNN) feature extractor to capture local physiological features.
• Evaluated the model on a dataset of 36,537 beats from 48 arrhythmia patients, using invasive arterial waveforms as the reference standard.

Main Results:

• BM-BPW effectively managed BP waveform variability in arrhythmia patients.
• Achieved low estimation errors: 0.04 ± 5.03 mmHg for systolic BP (SBP) and -0.05 ± 3.20 mmHg for diastolic BP (DBP).
• Demonstrated superior performance compared to state-of-the-art methods in cuffless BP estimation.

Conclusions:

• BM-BPW offers an accurate and efficient solution for cuffless continuous BP waveform monitoring in arrhythmia patients.
• The model's ability to handle BP variability caused by arrhythmia is a significant advancement.
• This technology holds substantial potential for improving clinical management and patient outcomes in the arrhythmia population.