Computer-aided Detection Fidelity of Pulmonary Nodules in Chest Radiograph
Nikolaos Dellios1,2, Ulf Teichgraeber1, Robert Chelaru2, Ansgar Malich2, Ismini E Papageorgiou2
1Department of Experimental Radiology, Institute of Diagnostic and Interventional Radiology, Jena University Hospital, Friedrich-Schiller University, Jena, 2Institute of Radiology, Suedharz Hospital Nordhausen gGmbH, Nordhausen, Germany
Date of Submission: 17-Aug-2016, Date of Acceptance: 23-Jan-2017, Date of Web Publication: 20-Feb-2017.
Aim: The most ubiquitous chest diagnostic method is the chest radiograph. A common radiographic finding, quite often incidental, is the nodular pulmonary lesion. The detection of small lesions out of complex parenchymal structure is a daily clinical challenge. In this study, we investigate the efficacy of the computer‑aided detection (CAD) software package SoftView™ 2.4A for bone suppression and OnGuard™ 5.2 (Riverain Technologies, Miamisburg, OH, USA) for automated detection of pulmonary nodules in chest radiographs. Subjects and Methods: We retrospectively evaluated a dataset of 100 posteroanterior chest radiographs with pulmonary nodular lesions ranging from 5 to 85 mm. All nodules were confirmed with a consecutive computed tomography scan and histologically classified as 75% malignant. The number of detected lesions by observation in unprocessed images was compared to the number and dignity of CAD‑detected lesions in bone‑suppressed images (BSIs). Results: SoftView™ BSI does not affect the objective lesion‑to‑background contrast. OnGuard™ has a stand‑alone sensitivity of 62% and specificity of 58% for nodular lesion detection in chest radiographs. The false positive rate is 0.88/image and the false negative (FN) rate is 0.35/image. From the true positive lesions, 20% were proven benign and 80% were malignant. FN lesions were 47% benign and 53% malignant. Conclusion: We conclude that CAD does not qualify for a stand‑alone standard of diagnosis. The use of CAD accompanied with a critical radiological assessment of the software suggested pattern appears more realistic. Accordingly, it is essential to focus on studies assessing the quality‑time‑cost profile of real‑time (as opposed to retrospective) CAD implementation in clinical diagnostics.
Min Jae Cha, Myung Jin Chung, Jeong Hyun Lee and Kyung Soo Lee (2019) Performance of Deep Learning Model in Detecting Operable Lung Cancer With Chest Radiographs. Journal of Thoracic Imaging34(2):86. doi: 10.1097/RTI.0000000000000388
Ju Gang Nam, Sunggyun Park, Eui Jin Hwang, Jong Hyuk Lee, Kwang-Nam Jin, Kun Young Lim, Thienkai Huy Vu, Jae Ho Sohn, Sangheum Hwang, Jin Mo Goo and Chang Min Park (2019) Development and Validation of Deep Learning–based Automatic Detection Algorithm for Malignant Pulmonary Nodules on Chest Radiographs. Radiology290(1):218. doi: 10.1148/radiol.2018180237
This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial-Share Alike 4.0 License, which allows others to remix, tweak, and build upon the work non-commercially, as long as the author is credited and the new creations are licensed under the identical terms.