Multi-Scale Convolutional Attention Enhanced Residual Network for Cuff-less Blood Pressure Prediction

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiovascular Monitoring

Background:

• Increasing demand for cuffless blood pressure monitoring in cardiovascular event management.
• Current research focuses on improving accuracy of non-invasive blood pressure devices using intelligent algorithms.
• Need for reliable cuffless methods to estimate systolic blood pressure (SBP) and diastolic blood pressure (DBP).

Purpose of the Study:

• To explore a novel cuffless method for estimating SBP and DBP.
• To utilize photoplethysmogram (PPG) and electrocardiogram (ECG) signals for blood pressure estimation.
• To develop an advanced deep learning framework for enhanced accuracy.

Main Methods:

• Proposed a novel residual network framework incorporating dynamic convolution and multi-scale convolutional attention.
• Employed dynamic convolution and channel adaptive mechanisms for low-level feature extraction from PPG and ECG signals.
• Utilized a parallel cross-mixing architecture with residual blocks and a multi-scale attention block (MSAB) for enhanced dependency modeling.

Main Results:

• Validated the framework on the VitalDB dataset, demonstrating excellent performance in SBP and DBP estimation.
• Achieved Mean Absolute Error (MAE) of 3.83 ± 5.71 mmHg for SBP and 2.29 ± 3.69 mmHg for DBP.
• The proposed framework outperformed existing models and met AAMI international blood pressure measurement standards.

Conclusions:

• The novel residual network framework provides accurate cuffless blood pressure estimation using PPG and ECG signals.
• The method's performance meets stringent international standards, indicating its clinical applicability.
• This approach represents a significant advancement in non-invasive cardiovascular monitoring technology.