Continuous Blood Pressure Dataset Featuring Arrhythmia and Diverse Baselines for Blood Pressure Estimation

Shuangdu Li, Ziyou Li, Yixuan Li, Xiaomao Fan, Yumeng Liu, Wenjun Ma, Bowen Zhang, Jianhua Ye, Ye Li

Abstract: Continuous blood pressure (BP) monitoring holds significant clinical value for the early diagnosis and prevention of cardiovascular conditions. Although arterial invasive lines remain the gold standard for BP assessment, they can cause discomfort and increase infection risk. Deep learning-based methods are widely employed for non-invasive continuous BP monitoring. However, their accuracy may be affected by arrhythmic conditions in physiological signals. This study aims to investigate the impact of rhythm changes on the predictive outcomes of various methods by constructing a novel continuous blood pressure dataset featuring arrhythmia called BP-ARR. Specifically, we constructed the BP-ARR dataset to include arrhythmic conditions and assessed the performance of established baseline methods under both normal and arrhythmic scenarios. Our comprehensive evaluation on the BP-ARR dataset reveals that the established baseline methods exhibit a decline in performance in BP estimation under arrhythmic scenarios compared to normal conditions, with an average increase in SBP MAE of 1.6 mmHg. Notably, the most significant increases were observed in ResUNet + Self-Attention (SBP MAE increased by 3.65 mmHg, 35.2%) and U-Net (SBP MAE increased by 2.21 mmHg, 40.8%).

External IDs:dblp:conf/adma/LiLLFLMZYL25