Wednesday, December 4 | 9:40 a.m.-9:50 a.m. | W3-SSIN06-02 | Room E450B
AI can assist radiologists in detecting acute cases of pulmonary embolism (PE), according to the results of a large, prospective study.
Researchers led by presenter Shlomit Goldberg-Stein of Northwell Health/Hofstra-Zucker School of Medicine sought to assess the concordance rate of radiologists with a commercial AI algorithm (Aidoc) for detecting acute PE in the clinical environment.
In a study involving over 32,000 consecutive CT exams with a PE protocol, they found that 3% of positive PE findings were detected only by the AI algorithm. However, 12.5% of positive PE cases were only detected by the radiologist.
Overall, the AI algorithm and radiologist agreed in 97.8% of the cases. Of the remaining cases that were not concordant, adjudicators sided with the radiologist’s interpretations in 77% of these exams. What’s more, the adjudicators agreed with the radiologist in 85% of the cases that had a positive diagnosis of PE.
“Radiologists should not be swayed against a positive diagnosis of PE when not detected by AI, given high concordance with adjudicators for positive PE (85%),” the authors wrote. “What remains unknown is the true added value of AI in detection of PE that would otherwise go undetected by radiologist since radiologists were aware of AI results prior to interpretation.”
Delve further into the findings by attending this talk on Wednesday morning.