AITRICS developed Deep Learning Model for real-time mortality prediction in critically ill children
October 22, 2019
AITRICS has developed a Deep Learning Model ‘PROMPT’ to predict risks for children’s intensive care units.
The paper, published in the world-renowned medical journal, Critical Care, is about a deep-learning model that shows the patient's risks in a time series through Vital Signs in intensive care units. The introduced deep learning model, conducted by Severance Hospital, Samsung Medical Center, and AITRICS as a joint research, quantifies the patient's condition including the risk of mortality in the intensive care unit. This enables medical team to make quick judgments and responses based on objective data.
In the pediatric intensive care unit, Pediatric Index of Mortality (PIM) or Pediatric Risk of Mortality (PRISM) were mainly used to predict mortality of patients. However, those algorithms had a major downside that they are not reflecting the patient conditions, because only fragmentary information in the early stages of childhood intensive care units predicts the risk of death.
The Pediatric Risk of Mortality Prediction Tool (PROMPT), developed in this study, analyzes vital signs, age, and weight of patients in the intensive care unit through deep learning to predict the risk of death from six to 60 hours. By using the Convolutional Deep Network, the time series of vital signs can be identified to significantly improve prediction accuracy.
AITRICS and Severance Hospital will later conduct follow-up studies that can improve the accuracy of models and ultimately reduce mortality in the pediatric intensive care unit by utilizing more data such as basic medical history information and diagnostic test results.
In addition to the deep learning model for predicting child risk, AITRICS is conducting joint research on adult patients with Severance Hospital and developed VitalCare, an in-hospital sepsis prediction solution, through the corresponding technology. With real-time monitoring through the patient's Electronic Medical Record (EMR), VitalCare significantly reduces the risk of developing sepsis in various environments in the hospital by timely detection of the risk factors. This improves hospital resource efficiency and health care services by supporting proactive response from the medical staff through accurate prior predictions of patient conditions in the hospital.
“It is an honor to have a deep learning-based pediatric intensive risk prediction model published in a global journal. The machine learning model developed in collaboration with AITRICS and Severance Hospital is recognized globally and is more meaningful. The use of the PROMPT model in an intensive care unit, where situations can change rapidly, can help prevent dangerous situations with proactive medical staff. The results of this study are expected to increase the survival rate of the pediatric intensive care unit” stated Saehoon Kim, Lead Research Scientist of AITRICS, who co-authored the study.
“PROMPT is an epitome of an effective intensive care unit solution where big data analysis and AI technologies can apply. I hope our work contributes to the improvement of the quality and efficient distribution of resources for intensive care treatments in the future” said Sooyeon Kim, professor of Severance Children's Hospital.
The paper, published in the Critical Care, is available on the AITRICS website (https://www.aitrics.com/publications/35/).