Objective:To evaluate the efficacy of artificial intelligence(AI)software in detecting pulmonary embolism(PE).Methods:We retrospectively analyzed chest CT pulmonary angiography(CTPA)images of 38 patients diagnosed with pulmonary embolism in the Medical Imaging Depart-ment of our hospital.After independent image review by AI software,38 patients were randomly divid-ed into two groups:Group A was reviewed by junior and intermediate-level physicians,and Group B was reviewed by junior and intermediate-level physicians with the assistance of AI.The diagnostic re-sults from three senior radiologists served as the gold standard for comparing the diagnostic timeliness and accuracy of PE between different review modes.Results:The sensitivity and specificity of the AI software in detecting pulmonary embolism were negatively correlated with the level of the pulmonary artery branch.For arterial branches of levels 1～-3,the diagnostic sensitivity was 100％,while the sensitivity was 92.05％,91.46％,and 83.33％,respectively,for levels 4,5,6,and above branches.The specificity for levels 1～2 branches was 100％,and for levels 3,4,5,6,and above branches,the specific-ity was 97.5％,96.43％,96.15％,and 96.15％,respectively.The diagnostic sensitivity(76.98％,92.85％)and specificity(89.81％,97.50％)of the junior and intermediate-level physicians assisted by AI were superior to those of manual review alone(62.42％,79.61％sensitivity and 79.67％,88.65％specificity).AI-assisted review reduced the reading time,and the highest diagnostic sensitivity and spe-cificity were observed with intermediate-level physicians assisted by AI(92.85％and 97.50％,respec-tively),with a significant reduction in reading time(P＜0.001).Conclusion:AI software for pulmonary embolism can improve the sensitivity and specificity of radiologists in diagnosing pulmonary embolism while shortening the reading time,and can be used as an effective AI-assisted tool in clinical practice.
Pulmonary embolismArtificial intelligenceComputed tomography pulmonary angiographyTomography,X-ray computedEffectiveness of the diagnosisSensitivitySpecificity
万方数据
维普
