Real-Time AI-Driven Monitoring of ICU Patients for Early Detection of Acute Respiratory Distress Syndrome (ARDS)

ARDS is a critical condition commonly encountered in intensive care units with a high death rate more than 40 %, and early diagnosis is very essential in patient outcome management. Current diagnostic approaches rely heavily on clinical judgment and static measurements, often resulting in delayed recognition and suboptimal treatment timing.The proposed work will design and construct an AI-based real-time monitoring system, which will predict the onset of ARDS 6-12 hours earlier than the patients may have shown their first clinical symptoms in the ICUs. We developed a deep learning ensemble model, consisting of convolutional neural networks and long short-term memory networks, that was trained using multimodal data such as continuous vital signs, laboratory data, and ventilator settings in addition to various types of imaging data of the chest over 2,847 ICU patients in three clinical centers. The model was validated using time-series cross-validation and external validation datasets. The AI system achieved 94.2% sensitivity and 89.7% specificity in ARDS prediction, with an average early detection time of 8.4 hours before clinical diagnosis, significantly outperforming traditional scoring systems. The implementation of this real-time monitoring system has the potential to revolutionize critical care by enabling proactive interventions, reducing mortality rates, and optimizing resource allocation in intensive care settings.

Hali tarjima qilinmagan