Researchers have developed an artificial intelligence-based patient monitoring system they say can identify signs of clinical deterioration earlier and more accurately than existing approaches. The system could help hospital staff intervene before a patient’s condition becomes critical. Details are discussed in the International Journal of Ad Hoc and Ubiquitous Computing.
Traditional monitoring systems rely largely on fixed thresholds for individual measurements such as heart rate, blood pressure or oxygen levels. However, these approaches often fail to account for differences between patients and may overlook how physiological changes interact across the body.
The new approach combines three machine-learning techniques. An adaptive attention mechanism continuously adjusts the importance assigned to different physiological signals. A spatiotemporal graph neural network analyses how vital signs influence one another and evolve. The system also incorporates reinforcement learning, a method in which algorithms learn decision-making strategies through feedback, enabling it to provide active clinical decision support rather than simply issuing alarms.
Tests were carried out to see how well the system performed in predicting historical outcomes recorded in two major intensive care unit (ICU) databases, MIMIC-III and eICU. The system achieved 96.3 per cent anomaly detection accuracy, generated warnings almost 40 minutes before critical events occurred, and reduced false alarms to 6.4 per cent.
Cheng, S., Zhu, J., Guan, S., Cheng, J. and Dou, T. (2026) ‘Intelligent monitoring of patient vital signs based on adaptive attention fusion spatiotemporal graph neural network’, Int. J. Ad Hoc and Ubiquitous Computing, Vol. 52, No. 5, pp.48–62.
No comments:
Post a Comment