Researchers at Changhua Christian Hospital and National Chung Hsing University deploy Random Forest and XGBoost models on Raspberry Pi edge devices to process ventilator-derived respiratory and pressure metrics, predicting extubation success and cutting server data uploads by over 80%, enhancing system reliability.

Key points

• Deployment of Random Forest and XGBoost on Raspberry Pi edge devices analyzing Vte, RR and airway pressures for extubation prediction.
• XGBoost outperforms Random Forest in tenfold and holdout validations, achieving over 90% accuracy with reduced inference time.
• Edge inference reduces server data uploads by 83.33%, minimizing latency and enhancing system stability for ICU decision support.

Why it matters: Deploying AI models directly on edge devices cuts latency and data load, offering clinicians faster, more reliable extubation decision support.