Artificial intelligence evaluation of computerized tomography images of patients suspected for pulmonary thromboembolism in the emergency department

In this study; a computer-aided algorithm created using artificial intelligence was used to detect pulmonary embolism (PE) with computed tomography pulmonary angiography (CTPA) images in the early period and to accurately localize the pulmonary embolism, which is an important cause of mortality in adult patients. In this retrospective methodological study, the data of 1948 patients who admitted to Hacettepe University Hospital, Emergency Medicine Department between January 01, 2016 and December 31, 2020 and were examined for CTPA with the preliminary diagnosis of PE were analyzed. After the images of the patients were recorded in DICOM format, the images were standardized and evaluated on the algorithm written in Python programming language. CTPA images of these patients was evaluated by the designed artificial intelligence algorithm for the presence and absence of PE; If PE is present, in terms of the anatomical level of the emboli (massive, main pulmonary, lobar, lobar+segmental, segmental, subsegmental). 51.4% of the patients included in the study were female (n=1002) and 48.6% were male (n=946). The median age of the patients was 61.50 years. Based on the radiologist's interpretation as the gold standard; In terms of segmental PE, the sensitivity of artificial intelligence analysis was 82.3%, the specificity was 93.5%, the positive predictive value was 29.5%, and the negative predictive value was 99.4%. In terms of main pulmonary PE, the sensitivity of the artificial intelligence analysis was 84.2%, the specificity was 99.7%, the positive predictive value was 80%, and the negative predictive value was 99.7%. In terms of massive PE, the sensitivity of artificial intelligence analysis was 84.6%, specificity 99.6%, positive predictive value 82.5% and negative predictive value 99.7%. False positivity rate in the artificial intelligence analysis of all patients participating in the study was calculated as 54.4% (n=276). According to artificial intelligence analysis, 44.1% (n=127) of false positive cases were in subsegmental PE groups. As a result, the use of artificial intelligence analyzes in emergency departments, where 24-hour radiologist interpretation cannot be provided, will provide great convenience to emergency physicians, especially in the early diagnosis of main pulmonary artery and massive PEs. Artificial intelligence analyzes need to be further developed for the diagnosis of PEs in smaller pulmonary artery branches, especially subsegmentary PE.