A deep learning algorithm proved capable of excluding a majority of CT scans that did not include coronary artery calcium scoring information, potentially eliminating the need for radiologists to manually identify such scans themselves.
Dutch researchers developed and trained their algorithm on images gathered from thousands of participants in a large trial focusing on cardiovascular disease (CVD) morbidity and mortality. After internally and externally validating their approach, the group was able to automatically spot 86% of all CT scans without valuable CAC information.
First author Leonardus B. van den Oever, with the University of Groningen in the Netherlands and colleagues, said such an algorithm would be beneficial in screening programs designed to identify patients with CVD, a leading global killer, earlier.
“The results show that our pipeline in a screening population might be used to exclude scans with no CAC without the risk of false negatives, and thus might be used to reduce the workload for radiologists in CAC screening,” the authors wrote, noting such screening programs would likely add a large number of scans for radiologists to interpret.
For their study, van den Oever and colleagues used 60 patient scans to train their algorithm, along with two separate sets of 50 CTs without CAC scoring and a 50-image CT set for both internal and external validation.
Of the 50 negative cases in the internal and external validation set, 62% and 86%, respectively, were classified correctly, with no false negatives.
“Assuming a prevalence for CAC of 60% in a screening population at elevated risk, deploying our model would allow for a direct CAC negative classification of 34 out of 100 scans,” the authors noted. “That implies a 34% reduction in the number of scans due for manual evaluation, and represents a considerable reduction in radiologists’ workload in such a screening setting.”
There were a number of limitations to their research, the group acknowledged, including a mere 100 participants in the external validation dataset. Although there were no false negatives in that group, more robust testing will be needed before real-world feasibility, which is currently underway, they said.
Read the entire pre-proof story in the European Journal of Radiology.
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.