Computer-assisted Diagnosis of Pulmonary Embolism in Multidetector Computed Tomography
M Zimmermann, M Das, CK Kuhl, S Keil
Hong Kong J Radiol 2017;20:115-20
DOI: 10.12809/hkjr1716402
Objective: To assess the performance of a computer-assisted diagnosis (CAD) software in detecting pulmonary embolism (PE) on multidetector computed tomography (MDCT).
Methods: MDCT angiography data of 100 consecutive patients with clinically suspected PE were retrieved. Diagnosis (initial read) had been made in consensus by two out of 10 radiologists. The original data were transferred to a CAD software, and two other experienced chest radiologists reviewed all CAD markings to determine the ground truth for CAD and the standard of reference (final diagnosis) in consensus. The markings were categorised as true or false positive. For true positive, the involved vessel segment was classified as lobar, segmental, and subsegmental. Sensitivity, specificity, and negative and positive predictive values of initial and CAD diagnoses were compared.
Results: In the initial read, 27 and 73 patients were positive and negative for PE, respectively. Based on the standard of reference, five (6.9%) of the 73 patients negative for PE were determined to be false negative and the remaining 68 were true negative. The CAD software marked 86 and 14 patients with and without pulmonary emboli, respectively. Based on the standard of reference, 26 were true positive, 57 were false positive, 11 were true negative, and 6 were false negative. The CAD software detected pulmonary emboli in five (6.9%) of 73 patients who were initially read as negative, but failed to detect pulmonary emboli in six (22.2%) of 27 patients who were initially read as positive. Respectively for the initial read and CAD software, sensitivity was 84% and 81%, specificity was 100% and 16%, positive predictive value was 100% and 31%, and negative predictive value was 93% and 65%.
Conclusions: The CAD software can be a second reader but should not be used as a stand-alone tool for diagnosis; all MDCT should be reviewed by a radiologist.
Authors’ affiliation:
M Zimmermann, M Das, CK Kuhl, S Keil: Department of Diagnostic and Interventional Radiology, RWTH Aachen University Hospital, Aachen, Germany
中文摘要
應用多排電腦斷層掃描輔助診斷肺栓塞
M Zimmermann, M Das, CK Kuhl, S Keil
目的：評估在多排電腦斷層掃描（MDCT）中應用電腦輔助診斷（CAD）軟件檢測肺栓塞（PE）的性能。
方法：回顧分析連續100例臨床疑PE患者的MDCT血管造影資料。由10名中的2名放射科醫師作出診斷共識（初次閱讀）。將原始數據傳輸到CAD軟件，並由另外兩名資深胸部放射科醫師審查所有CAD標記，以確定CAD的基本真實性和參考標準（最終診斷），並達成共識。標記被歸類為真或假陽性。對於真陽性，所涉及的血管段被分類為肺葉、節段和亞段。將初次和CAD診斷的敏感性、特異性、陰性和陽性預測值進行比較。
結果：在初次閱讀中，27例和73例患者分別為陽性和陰性PE。根據參考標準，73例陰性PE患者中有5例（6.9%）為假陰性，其餘68例為真陰性。CAD軟件分別標記了86例和14例患者有和無PE。根據參考標準，26例為真陽性，57例為假陽性，11例為真陰性，6例為假陰性。CAD軟件檢測到73例最初為陰性的患者中有5例（6.9%）有肺栓子，但CAD軟件在27例最初為陽性的患者中，有6例（22.2%）檢測不到肺栓子。初次閱讀和CAD軟件的靈敏度分別為84%和81%、特異度分別為100%和16%、陽性預測值分別為100%和31%、陰性預測值分別為93%和65%。
結論：CAD軟件可作為第二評估者，但不應被用作獨立診斷工具；所有的MDCT應由放射科醫師審查。