Revolutionizing Neonatal Care: Innovations in NICU Patient Monitoring | Carleton Newsroom

Advancing the NICU with AI 

One of the most promising aspects of Green’s research is the integration of artificial intelligence (AI) to interpret sensor data. By using machine learning algorithms, his team analyzes the context of the baby’s environment to differentiate between interventions, such as feeding, bathing, or diaper changes.  

“This information could change how clinicians interpret alarms and care for their tiny patients,” says Green.

To train a machine learning system, Green and his students manually observed patients in CHEO’s NICU that were using the pressure-sensitive mat and video sensor. They worked in collaboration with Dr. JoAnn Harrold, now-retired nurse Cheryl Aubertin, and retired Director of Clinical Engineering, Kim Greenwood.   

Green’s team observed 32 children from vulnerable premature infants to nearly ready-for-discharge babies. Spending an average of four hours collecting data on each patient, they often conducted their studies in the middle of the night to avoid disrupting parent visits. Using an application they developed for an iPad, they manually inputted each event of clinical significance such as sneezes and coughs, alarms and routine care events like diaper changes. 

“We wanted to capture all the events of clinical significance in real time to align it with the sensor data, so that later we can detect things like coughs and sneezes just from specific spikes on the pressure sensitive mat,” Green explains.   

Toyin Adams, fourth-year Carleton engineering student

Toyin Adams, a fourth-year Software Engineering student supervised by Green, is taking the integration of AI in the NICU one step further. With the information collected from the non-contact sensors, she is using open-source large language models – advanced computer programs that understand and generate human language – to develop a system that can send automated text messages to parents, clinicians, and researchers.  

Each message is tailored based on the audience needs. For example, a parent may receive a concise update on their baby’s condition, while a clinician would receive a more detailed report for handover rounds. 

“The potential benefits of automated message generation from non-contact sensor data are significant,” says Adams. “In critical environments like the NICU, where quick decisions are crucial, such summaries could be a game-changer for clinicians.”