Dataset of AABPE-RAG

Continuous monitoring of blood pressure (BP) is critical for the early diagnosis and treatment of cardiovascular diseases. Existing BP estimation methods mainly rely on mathematical equations or convolutional neural networks (CNNs) based on Photoplethysmography(PPG) features, suffering from two key limitations: over-reliance on single physiological indicators (failing to capture comprehensive medical information) and neglect of cardiovascular domain knowledge (compromising professional validity). To tackle these issues, we propose an AI Agent System Based on Retrieval-Augmented Generation about Multi-Physiological Fusion Continuous Blood Pressure Estimation(AABPE-RAG) which driven by Large language model(LLM), coupled with a wearable physiological signal monitoring bracelet to monitor PPG and Electrocardiography(ECG) signals. Leveraging LLMs\u2019 massive data analysis capability, we developed a domain-specific expert model via Prompt fine-tuning and established a cardiovascular knowledge base to mitigate hallucinations, and also enhance the professionalism of the model through the retrieval-augmented generation(RAG) technology. The model was trained on 50,000 samples from the MIMIC Database (83 physiological indicators) and evaluated on 18 datasets from 6 subjects. The mean absolute error \u00b1 standard deviation (MAE \u00b1 SD) for systolic BP (SBP) and diastolic BP (DBP) was 2.07 \u00b1 2.51 and 3.23 \u00b1 3.47, respectively.  This performance meets the standards of the Association for the Advancement of Medical Instrumentation (AAMI) and Grade A of the British Hypertension Society (BHS), outperforming current state-of-the-art approaches. Moreover, this study provides a new approach for AI consultation based on the real physiological data of subjects.