Gyrosphygmogram-Based Blood Pressure Estimation: A Comparative Study of Pulse Transit Time and CNN-LSTM Methods

Goal: Cuffless blood pressure (BP) monitoring is essential for continuous cardiovascular health assessment. This study evaluates the potential of a novel wearable signal, gyrosphygmogram (GSG), by comparing conventional pulse transit time (PTT)-based methods with a deep learning framework. Methods: ECG, PPG, and wrist-based GSG signals were simultaneously acquired from 20 healthy adults using Shimmer sensors, with the Omron 10 Series sphygmomanometer as the reference. PTT was computed across multiple timing pairs (ECG-PPG, ECG-GSG, and GSG-PPG) and modeled using linear and nonlinear regression equations. A multimodal deep learning network was developed to fuse temporal and morphological features from the three signals. Model performance was assessed against three different validation standards. Results: The deep learning model outperformed conventional regression, with GSG-ECG fusion yielding Mean Absolute Errors (MAEs) of 2.89 mmHg (systolic) and 2.2 mmHg (diastolic). Conclusions: Incorporating GSG into multimodal deep learning substantially improves cuffless BP estimation compared with traditional PTT. These findings highlight GSG's potential for enabling next-generation wearable devices with clinically validated accuracy.