Real-Time Prediction and Alarm System for Heart Rate and Blood Pressure: AI- for Anesthesia and Critical Care

An intelligent real-time prediction and alarm system that integrates deep learning with mode alarm strategies, improve the accuracy, timeliness, and clinical relevance of physiological monitoring. Traditional monitoring systems often rely on rigid threshold-based alarms that fail to account for dynamic biosignal fluctuations, leading to high false alarm rates and limited adaptability. The system extracts physiological data using optical character recognition (OCR) and standardizes the data to build a training dataset. A deep learning framework incorporating eight candidate models, including RNNs, CNNs, and Transformers, was established, with automatic model selection performed via Optuna and Pareto frontier analysis to identify the optimal configuration. To enhance prediction under limited data conditions, we proposed “pre-learning” and “self-learning” strategies and conducted a systematic analysis of signal volatility’s impact on generalization. The system supports two clinically oriented alarm modes: a fixed-baseline mode for stable conditions and a dynamic-baseline mode for detecting trends and drift. Additionally, a self-optimization module identifies 18 predefined false alarm features and recommends corresponding mitigation strategies through a closed-loop learning pipeline. By integrating multi-model deep learning, adaptive alarm logic, and self-optimization, we developed a flexible and intelligent physiological monitoring system. By tailoring alarm modes to clinical context and dynamically adjusting to signal characteristics, the system offers a reliable solution to enhance perioperative safety.