Abstract
Blood pressure (BP) estimation is critical for the accurate measurement and monitoring of BP. This study advances traditional methods by proposing a novel approach to estimate and predict BP using a graph-based network model. The model leverages connection information to depict the strengths and weaknesses of the relationships between signals, thereby enhancing the accuracy of BP estimation. We employed Euclidean distance (EDC) to express relationships between signals, whereas partial directed coherence (PDC) was used as a statistical method. In addition, our graph-based model used peak interval-based adjacency to represent these relationships in an adjacency matrix. Furthermore, this study aimed to create an accurate and reliable model using multichannel physiological signals. Model performance was evaluated using intra-subject models for individual assessments and inter-subject models across all subjects. The intra-subject evaluation of our peak interval-based adjacency model showed an mean absolute error (MAE) of 1.933, root-mean-squared error (RMSE) of 1.805, and correlation coefficient r of 0.515 for systolic blood pressure (SBP), and a MAE of 2.674, RMSE of 1.802, and r of 0.534 for diastolic blood pressure (DBP). The inter-subject evaluation demonstrated a MAE of 5.579, RMSE of 2.413, and r of 0.460 for SBP, and an MAE of 9.599, RMSE of 2.602, and r of 0.470 for DBP.
Youngshin Kang
PhD Student
Her research interests include lightweight deep learning algorithms, signal processing, and the Internet of Things.
Cheolsoo Park
Professor
His research interests include machine learning, adaptive signal processing, computational neuroscience, and wearable technology.