AI Case Study
PLA General Hospital in Beijing is using an algorithm for predicting if patients will wake up from a coma
The PLA General Hospital in Beijing has leveraged artificial intelligence to assist its doctors in evaluating whether patients will wake up from a come. The AI algorithm, developed by the Chinese Academy of Sciences, analyses fMRI brain scans to capture how blood flows to different areas of the brain. The algorithm's "diagnosis"
Industry
Healthcare
Healthcare Providers And Services
Project Overview
"The AI algorithm, developed by the Chinese Academy of Sciences and PLA General Hospital in Beijing, analyzes fMRI scans of a patient’s brains to gauge how blood flows to different areas of the brain, as well as information given by doctors like the patient’s age, how long they’ve lost consciousness, and the cause of the coma, and then makes its diagnosis. The algorithm and research underlying it were announced in the journal eLife in August 2018.
When the patients’ families were told the algorithm’s scoring, the patients’ doctors said not to base their entire decision to continue life support on the algorithm’s assessment. That’s because the algorithm isn’t right every time. (For example, a 36-year-old man who was scored to not recover by both doctor and algorithm ended up making a full recovery within a year.) Plus, the algorithm can only predict what’s happening inside the patient’s brain, meaning it couldn’t account outside factors like, say, a disease caught from another patient in the hospital. (Of course, a doctor wouldn’t be able to predict that either.)"
Reported Results
"The researchers claim the system is 88% accurate at predicting whether a patient would recover within a year. The researchers aren’t yet sharing results on how accurate the algorithm has been in the real world, but are now working to collect more data".
Technology
Function
R And D
Core Research And Development
Background
Benefits
Data
"Scans of 160 patients with consciousness disorders, meaning they were either in a vegetative state or minimally conscious. While this dataset is small compared to the hundreds of thousands of images typically used to train other image-based AI algorithms, the researchers claim the system is 88% accurate at predicting whether a patient would recover within a year.
Data analyzed by the algorithm came from two medical centers, which proved to have different kinds of patients, says co-author Tianzi Jiang. In addition, differences in the kind of scanner used and imaging protocols created variations in the two sets of data."