Neural network as accurate as board-certified radiologists, IDs some abnormalities they miss
TUESDAY, Oct. 22, 2019 (HealthDay News) — A deep learning algorithm has accuracy comparable to that of radiologists for the diagnosis of acute intracranial hemorrhage on head computed tomography (CT), according to a study published online Oct. 21 in the Proceedings of the National Academy of Sciences.
Weicheng Kuo, Ph.D., from the University of California Berkeley, and colleagues trained a fully convolutional neural network to detect neurologic emergencies (e.g., traumatic brain injury, stroke, and aneurysmal hemorrhage) utilizing 4,396 head CT scans performed at the University of California at San Francisco and affiliated hospitals. The algorithm’s performance was compared to the performance of four American Board of Radiology-certified radiologists on an independent test set of 200 randomly selected head CT scans.
The researchers found that the algorithm demonstrated the highest accuracy to date for this clinical application, with a receiver operating characteristic area under the curve of 0.991 for identification of acute intracranial hemorrhage. The algorithm exceeded the performance of two of the four radiologists. The algorithm detected some abnormalities missed by the radiologists. The algorithm’s high accuracy could perform pixel-level delineation of abnormalities to classify abnormalities into different pathological subtypes.
“We show that deep learning can accurately identify diverse and very subtle cases of a major class of pathology on this ‘workhorse’ medical imaging modality,” the authors write.
Copyright © 2019 HealthDay. All rights reserved.
