Clinical Deterioration Prediction in Smart Homes: A Sensor-Based Early Warning System Using Deep Learning

Background and Aims: Monitoring and predicting patient vital signs using Internet of Things technology in smart homes enables the early detection of clinical deterioration (CD). By analysing this data with artificial intelligence (AI), it is possible to prevent the consequences of CD. This study aims to develop an early warning score (EWS) system that predicts patient vital signs including heart rate (HR), blood pressure (BP), respiration rate (RR), body temperature (BT), and SpO۲ using unobtrusive sensors in smart home settings. Methods: We develop the AI and IoMT-based EWS system in three major phases. First, chair sensors in smart homes capture the study subjects’ BT and photoplethysmography (PPG) signals. In the second phase, we convert the PPG signals into vital signs, including HR, BP, RR, and SpO۲. Additionally, we develop a long short-term memory (LSTM) model to predict the five components of the EWS for the next four to six hours based on their previous five consecutive values. Finally, we create a web-based EWS system for our subjects based on their projected vital signs for four to six hours. Results: The mean absolute errors for the extracted vital signs were as follows: HR – ۲.۲۵ beats/min, RR – ۵.۲۸ breaths/min, SpO₂ – ۵.۱۱%, systolic BP – ۲.۸۱ mmHg, and diastolic BP – ۳.۱۰ mmHg. The LSTM model’s prediction results at the sixth time point almost align with the actual values for all vital signs. However, the body temperature prediction at the sixth time point shows the most significant deviation from its actual values. The calculated EWS for two participants was more accurate than for the others. The average EWS error based on the predicted vital sign values is ۱.۸۶ points. Conclusion: Our results demonstrate the potential of using unobtrusive sensors and AI algorithms to remotely monitor discharged patients in their homes. However, there is a need to implement more advanced denoising techniques for collecting PPG signals.

کلیدواژه ها:

Artificial Intelligence ، Clinical Deterioration ، Smart Home