A Bidirectional Long Short-Term Memory Deep Learning Model for Classification of Pulse Waveform

Area of Science:

• Biomedical Engineering
• Cardiovascular Physiology
• Artificial Intelligence in Medicine

Background:

• Arterial blood pressure (ABP) waveform morphology is a key indicator of cardiovascular status.
• Existing wave separation analysis (WSA) methods often rely on invasive measurements and are limited to physiological waveform types.
• Identifying altered vascular compliance and resistance through non-invasive means remains a challenge.

Purpose of the Study:

• To develop and validate a deep learning model for classifying arterial blood pressure (ABP) beats into distinct morphological types (Type A vs. Type B/C).
• To assess the model's performance using both central (aortic) and peripheral (femoral) ABP waveforms.
• To contribute to enhanced non-invasive cardiovascular monitoring and early detection of arterial alterations.

Main Methods:

• Implementation of a bidirectional long short-term memory (BiLSTM) deep learning architecture.
• Training and testing the BiLSTM model on datasets of central and peripheral ABP waveforms.
• Classification of ABP beats into Type A (physiological) and Type B/C (altered vascular compliance/resistance).

Main Results:

• The BiLSTM model achieved high classification accuracy: 96% for aortic waveforms and 90% for femoral waveforms.
• The model effectively distinguished between normal and altered ABP waveform morphologies.
• Demonstrated the feasibility of using deep learning for ABP waveform classification.

Conclusions:

• Bidirectional LSTM models can accurately classify ABP waveform morphology from both central and peripheral signals.
• This deep learning approach offers a promising non-invasive method for assessing vascular conditions.
• The findings support the potential for improved early detection of arterial alterations and cardiovascular monitoring.