- Four Institutions Join Free Healthcare AI Pilot for Radiologists
“We worked in collaboration with GE Healthcare, and the goal was to develop a platform called the Critical Care Suite that would detect critical findings on chest radiographs,” Yaeger told HITinfrastructure.com.
Yaeger and his team cooperated with an international team of researchers in developing the AI-powered x-ray machine, which was submitted to the Food and Drug Administration for review last November.
AI Enabled Prioritized Review of Chest X-Rays
“By having the AI platform running directly on chest x-ray system, this enables prioritized review if the algorithm detects a pneumothorax. It would alert the radiologist that there’s a positive finding. Then we hope to take interpretation and turnaround times down to the order of minutes, as opposed to potentially more than an hour depending on when a chest x-ray is taken,” Yaeger said.
“The first step of the development process was to prioritize the list of findings that we wanted to detect. We started with pneumothoraces, which are potentially life-threatening and require emergency attention and treatment,” Yaeger explained.
“To train the system, first we started with a set of tens of thousands of x-rays that had to be reviewed for both true positive and true negative findings,” he said.
“The training set was annotated by multiple radiologists at multiple institutions. That was then used to train the AI algorithm. We conducted a reader study using three radiologists to establish ground truth and compared the radiologists' performance to the AI algorithm performance,” he related.
74,000 Americans Suffer Collapsed Lungs Annually
Some 74,000 Americans suffer collapsed lungs each year, resulting from trauma, cigarette smoking, drug abuse, and certain medical conditions.
Yaeger explained that they decided to use AI for adaptability and the ability to continually improve performance.
“We are in no way trying to replace the radiologist for review. We're trying to prioritize the review and also call attention to areas of interest so the radiologist can see what was concerning to the algorithm, essentially a beacon that says 'Look here, double check this area for pneumothorax'," he said.
Yaeger sees a bright future for AI applications in healthcare. In terms of radiology, he sees a potential use of AI for detecting intracranial hemorrhage, various types of fractures, and certain types of early stage cancers.
Outside of radiology, he believes that a big advantage of AI is interpreting large data sets, such as data mining multi-institutional databases to uncover hidden diseases among the population.