A Pilot Study on Automatic Detection of Pulmonary Embolism Using Fractional Pulmonary Defect Area Derived from...

This study proposed a processing algorithm of dynamic chest radiography (DCR) for diagnosing pulmonary embolism (PE) based on perfusion defect area in the image of lung perfusion (LP) map with optimal perfusion defect criteria (PDC). Seventy-seven patients (16 with acute PE and 16 with chronic PE) who underwent DCR and chest contrast-enhanced computed tomography within 72 h of the interval were retrospectively enrolled. PDC was set as the threshold for determining the presence or absence of perfusion defects based on the imported LP map. Twenty steps of PDCs were set, from - 5 to - 290, with a tolerance of - 15. The perfusion defect area was defined as the number of pixels exceeding the PDC in the right and left lung. The fractional perfusion defect area (FPDA) was calculated as its percentage relative to the total lung area. Sensitivity and specificity of PE diagnosis by FPDA at each PDC were calculated using receiver operating characteristic (ROC) analysis. PE detection by FPDA at the best PDC of - 110 and - 125 in right and left lungs showed cut-off values of 46.0% and 55.4%. Diagnostic performance including the area under the ROC curve (AUC), sensitivity, and specificity, respectively, along with 95% CIs, was as follows: right lung-88% (0.80-0.97), 0.75 (0.58-0.87), and 0.89 (0.77-0.95)-and left lung-81% (0.71-0.91), 0.56 (0.39-0.72), and specificity 0.93 (0.82-0.98). FPDA was proposed for automatic PE diagnosis using a DCR-based LP map. FPDA, derived from non-invasive DCR, provides objective and accurate additional information in the initial flow of PE diagnosis.