An artificial intelligence (AI) algorithm was applied to a retrospective cohort of 1,387 consecutive CT pulmonary angiograms performed at a large academic institution between Jan. 21 and April 28, 2019. Natural language processing (NLP) of the cohort reports was used to identify if PE was present. Discrepant cases between the algorithm and NLP results were reviewed by three board-certified thoracic radiologists and the emboli called in error were graded based on location. Quality of contrast bolus and respiratory motion were also graded.
The prevalence of PE was 13.6% (189 cases). The algorithm was 93% sensitive and 96% specific in the detection of PE. The positive predictive value was 77% and the negative predictive value was 99%.
Our AI algorithm is sensitive and specific in the detection of PE on CTPA. The high negative predictive value suggests this algorithm has the potential of becoming a screening tool to aid expedite diagnosis and appropriate management of patients with pulmonary embolism.