The AI Will (I)CU Now: Deep Learning Helps Guide Decisions in Intensive Care (Image credit- Nvidia)
Researchers from the Human-Computer Interaction Institute (HCII) at Carnegie Mellon University, the University of Pittsburgh, and UPMC conducted a study that suggests AI could help clinicians make quick and difficult judgments, particularly in intensive care units.
The researchers unveiled an interactive clinical decision support (CDS) interface called the AI Clinician Explorer that uses an AI Clinician model to suggest treatments for sepsis.
When the body reacts to an infection by damaging its own tissues and organs, it develops sepsis, a potentially fatal condition.
It can happen when the immune system’s molecules that are meant to fight infection instead induce inflammation throughout the body.
More than 18,000 patients who satisfied the established diagnostic standards for sepsis throughout their stays in the ICU served as the model’s training data.
Now that the dataset has been filtered and searched, medical professionals may see the progression of the patient’s diseases and contrast the predictions generated by the model with the actual treatment choices made at the bedside.
24 ICU doctors with experience treating sepsis used a condensed AI Clinician Explorer interface to decide how to handle four fictitious patient scenarios in a think-aloud study.
Based on their actions, the researchers divided the doctors into four groups: ignore, rely, consider, and negotiate.
The “rely” group regularly followed the AI’s recommendations, but the “ignore” group constantly ignored them.
Before adopting or rejecting the AI’s recommendation, the “consider” group assessed it. The vast majority of participants belonged to the “negotiate” group, which partially adopted the AI’s suggestion but rejected others.
The research team found that the majority of clinicians used the AI Clinician model in some of their choices. As a member of the study team and a Ph.D. student at the HCII, Venkatesh Sivaraman thinks that although physicians are enthusiastic about how AI may be able to assist them, they might not be familiar with how these AI technologies operate.
The objective of the research is not to supplant or mimic the judgments of clinicians. Instead, it uses AI to find trends in patient outcomes that may have gone unnoticed in the past. The team asserts that the system might direct therapists in a different route or support their existing course of action.
However, some medical professionals expressed concern about how little access the AI had to comprehensive patient data, such as general appearance. When the AI proposed methods that were different from what they had been taught, they were likewise dubious.
Also read: Volvo and CERN are collaborating to increase the safety of autonomous vehicles
“There was not a good sense of why the CDS deviates from what clinicians would normally do or consider to be best practice,” Sivaraman said in a statement for a news release.
We are currently concentrating on figuring out how to supply that data and evaluate these recommendations because it is a difficult subject that will involve machine learning and AI.
