The Radiological Society of North America and partners have assembled the largest-ever annotated collection of brain hemorrhage CT images, making it available to artificial intelligence developers looking to tackle this topic.
RSNA announced this “unprecedented collaboration” on Wednesday, which was made possible with the help of 60 volunteer neuroradiologists and a fellow imaging interest group. The work occurred through the latest iteration of the society’s AI challenge, with data coming by way of the Federal University of São Paulo, Stanford and Thomas Jefferson’s Pennsylvania hospital.
“The value of this challenge is to create a dataset that might lead to a generalizable solution,” Adam Flanders, MD, neuroradiologist and professor at the Philadelphia-based institution, said in an RSNA statement. “The best way to do that is to train a model from data originating from multiple institutions that use a variety of CT scanners from various manufacturers, scanning protocols and a heterogeneous patient population.”
Alongside the American Society of Neuroradiology, the groups worked to curate the dataset, soliciting help from ASNR’s membership. Sixty volunteers stepped up to help annotate more than 874,000 brain hemorrhage images, labeling each as normal or abnormal.
After making the data publicly available, RSNA said it has received widespread interest, with 22,000-plus submissions from AI developers in 75 countries. Experts plan to use a similar model for this year’s AI challenge—in concert with the Society of Thoracic Radiology—targeting pulmonary embolism on chest CT.
“I was really impressed by the huge volunteer effort and the tremendous worldwide interest in this project. The dataset we created for this challenge will endure as a valuable [machine learning] research resource for years to come,” added Flanders, who is also lead author of an analysis published Wednesday in Radiology: Artificial Intelligence.