If it wasn’t for a mysterious hot pepper allergy, Harini Suresh might never have found a way to improve patient care in intensive care units.
Suresh, a doctoral student at MIT, wants to use AI to help critical care doctors choose the best treatment for each patient. That’s not easy when patients are sick with dire conditions like heart failure or stroke, and doctors must quickly weigh vast and varied patient data that may range from simple demographics to complex lab tests.
“The ICU is a high-stakes, high-demand environment, and doctors can spend only a limited amount of time with each patient,” said Suresh. “When doctors are dealing with many data sources and data types, computational tools can make a difference.”
How Peppers Led to Predictions
Suresh, who works in the Computer Science and Artificial Intelligence Lab (CSAIL) at MIT, was a girl when her mother, who’d happily enjoyed hot peppers all her life, developed a sudden allergy to their fiery spice. That sparked an interest in medicine that today drives Suresh’s research to make medicine more personal, starting with the ICU.
Her latest paper shows how GPU-accelerated deep learning predicts whether patients will need certain ICU treatments. The model uses hourly measurements of vital signs — such as blood pressure, heart rate and glucose levels plus patient information like age and gender, to forecast needed treatments.
Dubbed ICU Intervene, it can help doctors decide whether patients will need treatments to help them breathe, improve heart function, raise blood pressure or replace lost blood. It also predicts when patients can safely be taken off these treatments. So far, it can forecast eight hours in advance.
ICU Intervene is unique among deep learning solutions because it explains the reasoning behind its decisions.
“That (explanation) may be the most important factor in getting technology like this into use because doctors and patients need to have confidence in the decisions,” Suresh said.
How the Computer Learned Medicine
Suresh and the MIT research team — including Suresh’s advisor, MIT professor Peter Szolovits, and an MIT physician — trained ICU Intervene to associate factors like heart rate, oxygen saturation and kidney function with recommended treatments. For training data, the team took information about 34,000 ICU stays included in the MIMIC-III public database of de-identified patient health information.
They sped up training with NVIDIA GeForce TITAN X GPUs and cuDNN, then used the same GPUs to deploy their model.
If put into practice, ICU Intervene could help doctors more quickly choose the right treatment for each patient, Suresh said. It could also avoid unnecessary treatments.
Next, researchers will work to improve ICU Intervene to be able to give more individualized care and provide more advanced reasoning for decisions.
“If we can improve the quality of clinical decision-making, we could really improve the quality of patient care,” Suresh said.