ACİL SERVİSTE İSKEMİK SEREBROVASKÜLER OLAY ŞÜPHESİ TAŞIYAN HASTALARIN BİLGİSAYARLI TOMOGRAFİ GÖRÜNTÜLERİNİN YAPAY ZEKA İLE DEĞERLENDİRİLMESİ
Özet
In this study, the performance of an artificial intelligence-based computer software, that was created by researchers in the present study, in recognizing ischemia causing thrombi using brain-neck computerized tomography angiography images and in detecting patients suitable for endovascular treatment among patients who admit to the emergency department with preliminary diagnosis of ischemic cerebrovascular event, was evaluated. In this retrospective study, the data of 1860 patients who admit to Hacettepe University Faculty of Medicine Emergency Department between the dates 01.08.2017-31.07.2022 and who were administered brain-neck computerized tomography angiography imaging with preliminary diagnosis of ischemic cerebrovascular event were analyzed. After the images were downloaded in DICOM format, images were evaluated uniformly using the algorithm that was written with Python coding language. Brain-neck computerized tomography angiography images were evaluated in terms of presence or absence of a thrombus using the artificial intelligence-based algorithm that was developed by the researchers. 48.2% of the patients were female (n=897), 51.8% were male (n=963). Median age was 64.65 years. Performance parameters of the algorithm on detecting thrombi when radiologist interpretation was defined as the gold standard were as follows: sensitivity 62.89%, specificity 19.13%, positive predictive rate 19.3%, negative predictive rate 76.11%. Parameters of the algorithm on detecting thrombi on ICA were calculated as: sensitivity 92.9%, specificity 22.2%, positive predictive rate 2.14%, negative predictive rate 99.19%. Parameters of the algorithm on detecting thrombi on MCA were calculated as: sensitivity 80.7%, specificity 21.9%, positive predictive rate 5.19%, negative predictive rate 99.55%. Parameters of the algorithm on detecting thrombi on PCA were calculated as: sensitivity 90.5%, specificity 22%, positive predictive rate 0.02%, negative predictive rate 99.19%. Parameters of the algorithm on detecting thrombi on ACA were calculated as: sensitivity 88.9%, specificity 21.9%, positive predictive rate 0.90%, negative predictive rate 99.59%. In conclusion, the use of artificial intelligence-based algorithms in emergency departments where radiology consultation is not available 24/7, will be greatly helpful for practitioners in early detection or exclusion of cerebrovascular thrombi. Although the software created in the present study needs to be further developed and updated in many aspects, it can be considered as a preliminary screening test in the clinical practice.