Overview
Research focused on continuous and cuffless blood pressure (BP) monitoring using photoplethysmography (PPG) identified challenges stemming from population heterogeneity and the 'one-to-many mapping' problem, where similar PPG waveforms correspond to differing BP levels across individuals. To address these issues, a novel dual-path architecture, termed SIFPBPNet, was proposed. This network is designed to separately represent and integrate steady-state and instantaneous features for improved BP estimation.
Research Context
The development of continuous and cuffless BP monitoring from PPG signals holds significance for personalized cardiovascular health management due to its potential for low-cost application. Existing population-based models face limitations in accuracy, which is attributed to significant population heterogeneity. A key identified challenge is the 'one-to-many mapping' problem, implying that similar PPG waveforms do not consistently map to the same BP levels across different individuals.
Approach
The SIFPBPNet architecture incorporates two distinct pathways: a Steady-state Feature Path (SFP) and an Instantaneous Feature Path (IFP). The SFP is designed to extract individual-specific and long-term characteristics. It utilizes a Graph Attention Network (GAT) to process multi-day historical PPG trajectories, thereby representing the steady-state features of an individual. Concurrently, the IFP is tasked with capturing short-term dynamics. It processes current PPG segments and integrates the steady-state prior information derived from the SFP through a cross-attention mechanism.
Findings
- The SIFPBPNet architecture demonstrated specific performance metrics for blood pressure estimation. On a large-scale wearable dataset, it achieved a Mean Absolute Error (MAE) of 8.57 mmHg for systolic BP and 5.97 mmHg for diastolic BP.
- These results indicated that SIFPBPNet outperformed state-of-the-art models when evaluated on the same dataset.
- Further analysis revealed that the Steady-state Feature Path (SFP) module consistently improved performance when integrated into various backbone architectures.
- Specifically, the integration of the SFP module yielded relative MAE reductions ranging from 2.8% to 13.1% for systolic BP.
- These findings collectively suggest strong generalizability and plug-and-play transferability of the SFP module within different model frameworks.
Why This Matters
The SIFPBPNet model and its SFP module address challenges in cuffless BP monitoring by explicitly accounting for individual physiological differences and long-term trends, which can enhance the accuracy of personalized cardiovascular health management.