Evaluation of artificial intelligence detection of pulmonary embolism in dual-energy computed tomography

Dual-energy CT can be utilized on CT pulmonary angiography (CTPA) exams, increasing detection of peripheral PE. Artificial Intelligence (AI) software has emerging applications including the detection of PE; however, these AI models are not exclusively trained on dual-energy data. This study aims to evaluate the diagnostic performance of AI PE detection software on dual-energy virtual monochromatic (VMC) series compared to single-energy CTPA exams.

A retrospective study analyzing 200 CTPA exams comprised of dual-energy and single-energy studies with an equal distribution of scans reported as positive and negative for PE creating four groups for analysis. Dual-energy studies consisted of 50 keV and 70 keV VMC series. Each case was analyzed by the Aidoc pulmonary embolism detection software. In cases with a discrepancy between the radiologist and the AI detection software, two blinded radiologists reviewed the scan to establish a majority truth. Sensitivity and specificity analyses was calculated for each of the four groups.

AI PE detection software had an 84.6 % sensitivity and a specificity of 100 % for single-energy PE exams. For the 50 keV dual-energy VMC PE cases the AI software had an 84.3 % sensitivity and a specificity of 100 %. For the 70 keV dual-energy VMC PE cases the AI software had an 87.5 % sensitivity and 100 % specificity. No significant difference was seen in sensitivity or specificity based on overlapping 95 % confidence intervals of groups.

AI PE detection software can be utilized on dual-energy VMC CTPA exams without a significant difference in sensitivity of specificity when compared to single-energy CTPA exams.