Potential autonomous reporting rate was 7.8 percent for all posteroanterior chest radiographs
By Elana Gotkine HealthDay Reporter
WEDNESDAY, March 8, 2023 (HealthDay News) — An artificial intelligence (AI) tool has very high sensitivity for abnormal chest radiographs, according to a study published online March 7 in Radiology.
Louis L. Plesner, M.D., from Herlev and Gentofte Hospital in Copenhagen, Denmark, and colleagues conducted a retrospective study involving consecutive posteroanterior chest radiographs from adult patients in four hospitals obtained in January 2020. The radiographs were labeled by three thoracic radiologists as critical, other remarkable, unremarkable, or normal. Chest radiographs were classified using an AI tool as high confidence normal or not high confidence normal (abnormal).
A total of 1,529 patients were included for analysis: 1,100 (72 percent) were classified as having abnormal radiographs (including 617 [40 percent] classified as having critical abnormal radiographs) and 429 (28 percent) were classified as having normal radiographs. The researchers found that the sensitivity of AI was 99.1 and 99.8 percent, respectively, for abnormal radiographs and critical radiographs compared with sensitivities of 72.3 and 93.5 percent, respectively, for radiologist reports. The specificity, representing the potential autonomous reporting rate, of AI was 28.0 percent for all normal posteroanterior chest radiographs or 7.8 percent of all posteroanterior chest radiographs.
“The most surprising finding was just how sensitive this AI tool was for all kinds of chest disease,” Plesner said in a statement. “In fact, we could not find a single chest X-ray in our database where the algorithm made a major mistake. Furthermore, the AI tool had a sensitivity overall better than the clinical board-certified radiologists.”
Two authors disclosed financial ties to the medical device industry.
Editorial (subscription or payment may be required)
Copyright © 2023 HealthDay. All rights reserved.