Maintaining Blood Oxygen Saturation Level with a Deep Learning Neural Network Model Using Real-Patient Data

In times of pandemics, mechanical ventilators are indispensable for delivering controlled inspired oxygen (FiO2) to maintain appropriate blood oxygen saturation levels (SaO2). Abrupt drops in SaO2 can precipitate hypoxemia, leading to severe organ damage or even death. To address this challenge, AI-driven systems have emerged, mimicking physician decision-making processes to alleviate healthcare professional burdens and optimize critical care systems with timely responses. This paper presents the development of a mechanical ventilator advisory system using deep learning neural network algorithms. Leveraging real-patient data, the model accurately predicts FiO2 levels based on real-time patient parameters, enabling proactive interventions. Through extensive analysis, the model achieved an accuracy error below the acceptable threshold of 5%. This deep learning model proves invaluable, particularly in epidemic scenarios like COVID-19, offering vital support to healthcare providers in optimizing patient care. Future improvements involve expanding datasets and incorporating additional ventilator and patient parameters using deep neural networks. Furthermore, integrating ventilator modes and positive end-expiratory pressure (PEEP) as supplementary system outputs holds promise for enhancing accuracy and clinical utility. Practical implementation entails synchronizing ventilator prototypes with real-patient physiological parameters through standardized software and reliable data acquisition devices, further advancing patient-centered care in critical care settings.