Hospital, lab collaborate on sepsis-detection AI

EARLY WARNING:A system introduced last year can predict sepsis about four hours before its onset, but with AI, the hospital hopes to automatically detect it earlier

By Lee I-chia  /  Staff reporter



Taipei Medical University Hospital superintendent Chen Ray-jade, left, and Taiwan AI Labs founder Ethan Tu on Dec. 7 hold up a collaboration agreement introducing artificial intelligence at the hopsital in Taipei.

Photo: Wu Liang-yi, Taipei Times

Taipei Medical University Hospital on Dec. 7 signed a contract with Taiwan AI Labs to collaborate on introducing an artificial intelligence (AI) system that would provide an earlier warning for sepsis in its intensive care units (ICUs).

The five-year contract with the research organization was signed by hospital superintendent Chen Ray-jade (陳瑞杰) and Taiwan AI Labs founder Ethan Tu (杜奕瑾).

The first phase of the collaboration aims to improve the hospital’s “electronic dashboard” ICU (TED-ICU) by integrating an AI system that could automatically predict the early development of sepsis and alert the resident medical team.

The health conditions of ICU patients can change drastically, and sepsis is a life-threatening condition that occurs in about 437 patients per 100,000 globally, Chen said, adding that studies suggest that sepsis causes about one-third of all deaths in US hospitals.

Previously, sepsis could only be detected after its onset, but after the hospital last year introduced the TED-ICU, medical teams can predict the risk of sepsis about four hours before it occurs, Chen said.

The system works by automatically collecting, integrating, analyzing and keeping a record of ICU patients’ physiological information, the hospital said.

The collaboration with Taiwan AI Labs aims to further improve the dashboard with AI, allowing a real-time alert as soon as it detects the risk of sepsis, hopefully even more than four hours in advance.

Tu said that after his mother passed away from sepsis last year, he began thinking about how to use AI to interpret real-time physiological information to reduce the sepsis mortality rate.

The lab in September developed an AI sepsis detection system and allowed it to learn from the hospital’s data for two months, Tu said, adding that so far, its detection accuracy is at about 85 percent and can hopefully enter clinical use soon.

The hospital said it hopes to introduce AI to learn the medical records of people injured in falls to develop a mechanism to predict high-risk groups for such injuries and to analyze their causes to improve care.

source: Hospital, lab collaborate on sepsis-detection AI