Research Terms
Keywords
Applied Machine Learning Health And Clinical Informatics Natural Language Processing Smart And Connected Health
Industries
Intelligent Critical Care Center (IC3)
| Director |
Azra Bihorac Parisa Rashidi |
| Phone | 352-273-9009 |
| Website | https://ic3.center.ufl.edu/ |
| Mission | The Intelligent Critical Care Center will advance our vision for transformative critical and acute care medicine where pervasive and immersive artificial intelligence augments a human-centered healthcare system. The Center’s mission is to develop and provide sustainable support and leadership for transformative medical AI research, education, and clinical applications to advance patients’ health in critical and acute care medicine. Translational research in medical AI requires diversely trained faculty engaged in a transdisciplinary approach that creates a unity of intellectual frameworks beyond the disciplinary perspectives to ensure a trustworthy and fair evolution of these powerful tools towards patient care. Through the partnership between the University of Florida (UF) Colleges of Medicine (COM) and Engineering (COE), our Center is the first transdisciplinary program in the nation to study AI in critical and acute illnesses and provide the necessary collaborative infrastructure to propel the University of Florida and UF Health as the leaders in this field. |
This intelligent patient monitoring system uses an array of wearable sensors and cameras combined with machine-learning platforms to conduct autonomous, continuous observation and analysis of patient status and environmental conditions in the intensive care unit (ICU) or assisted living facility. The market for patient monitoring systems should exceed $25 billion by 2023. In critical care environments, patients require extensive monitoring of their physiological state to inform treatment decisions and ensure proper recovery. While much of a patient’s assessment in the ICU happens automatically through various bedside devices, available monitoring systems do not automatically quantify certain relevant environmental factors, such as noise or light levels, frequency of sleep-disrupting visitations, and patient delirium. Other aspects of ICU care, such as assessing pain or physical functionality, require manual evaluations, which are quite subjective, lack precision and increase the already overburdened hospital staff workload. Additionally, the intermittent nature of manual observations can inhibit medical interventions at critical times.
Researchers at the University of Florida have developed an autonomous ICU monitoring system that captures important patient information and environmental conditions and characterizes the physiological state of patients. The system uses a network of sensors and visual criteria to monitor and automatically analyze data such as a patient’s facial expressions, visitation frequency, and the room’s light and noise levels, providing continuous and automatic quantification of ICU factors for more timely medical interventions.
Autonomous ICU patient monitoring system that continuously collects and analyzes pervasive input and physiological information in real-time
This autonomous patient monitoring system employs pervasive sensing and machine learning procedures to evaluate an ICU patient’s environment and physiological status. A network of wearable sensors, room light and noise sensors, and high-resolution cameras continuously capture data on a patients and their environment in the ICU. Machine learning and statistical software analyzes the visual information, such as patient posture, expressions, and physical movements. In conjunction with the gathered environmental conditions and with reference to the patient’s vital signs and electronic health record, this automated analysis enables the system to evaluate the physiological state of a patient, including quantifying pain levels or identifying conditions such as delirium.
This deep learning model provides a continuous acuity score for critically ill patients in the intensive care unit (ICU), enabling earlier clinical interventions and improving patient outcomes. Patients experiencing a medical event may have a life-threatening condition or propensity to develop one at any moment. Accurate, timely assessment of patient acuity is essential in the ICU, where rapid changes in clinical status can have life-threatening consequences. Early recognition of evolving illness severity in critically ill patients is invaluable, helping identify patients in need of life-saving interventions and informing shared decision-making process among patients, providers, and families about goals of care and optimal resource utilization. Traditional scoring systems like SOFA rely on static, intermittent calculations and manual data entry, often missing subtle or sudden deterioration and delaying medical response. These limitations can lead to missed opportunities for early intervention and increased burden on clinical staff, who must synthesize large volumes of physiologic data while managing complex care. The demand for advanced patient monitoring solutions is rising, with the global market for patient monitoring devices projected to reach $71.1 billion by 2029.
Researchers at the University of Florida have developed a deep learning–based acuity scoring model that autonomously and continuously analyzes physiologic and patient background data in real-time for ICU patients. By leveraging advanced temporal deep learning models, the system generates dynamic risk assessments and mortality predictions. Unlike static scores calculated every 12–24 hours, this system updates continuously and autonomously without manual data entry. This approach delivers more accurate, interpretable insights to support earlier, data-driven interventions, ultimately improving patient outcomes and reducing the workload on ICU staff.
Continuous ICU acuity scoring system that autonomously analyzes real-time physiologic and patient background data to deliver dynamic risk assessments and mortality predictions, supporting timely, data-driven clinical interventions
This predictive model applies deep learning to generate continuous acuity scores for ICU patients by combining recurrent neural networks (RNNs), gated recurrent units (GRUs), and self-attention mechanisms. The system ingests a comprehensive array of physiologic measurements (blood pressure, oxygenation, urine output, lab values, vasopressor use, mechanical ventilation status) along with patient background information (demographics, comorbidities). RNNs learn patterns in patient data over time, GRUs optimize information flow for efficiency and stability, and self-attention mechanisms identify the most clinically significant time points. This approach enables accurate, interpretable risk predictions tailored for complex ICU environments.
