Purpose: To retrospectively determine interobserver variability of semiautomated volume measurements of pulmonary nodules and the potential reasons for variability.
Materials and methods: The Dutch-Belgian lung cancer screening trial (NELSON) is a lung cancer screening study that includes men between the ages of 50 and 75 years who are current or former heavy smokers. The NELSON project was approved by the Dutch Ministry of Health and the ethics committee of each participating hospital. Informed consent was obtained from all participants. For this study, the authors evaluated 1200 consecutive low-dose computed tomographic (CT) scans of the chest obtained during the NELSON project and identified subjects who had at least one 50-500-mm(3) nodule. One local and one central observer independently evaluated the scans and measured the volume of any detected nodule by using semiautomated software. Noncalcified solid nodules with volumes of 15-500 mm(3) were included in this study if they were fully surrounded by air (intraparenchymal) and were detected by both observers. The mean volume and the difference between both measurements were calculated for all nodules. Intermeasurement agreement was assessed with the Spearman correlation coefficient. Potential reasons for discrepancies were assessed.
Results: There were 232 men (mean age, 60 years; age range, 52-73 years) with 430 eligible nodules (mean volume, 77.8 mm(3); range, 15.3-499.5 mm(3)). Interobserver correlation was high (r = 0.99). No difference in volume was seen for 383 nodules (89.1%). Discrepant results were obtained for 47 nodules (10.9%); in 16 cases (3.7%), the discrepancy was larger than 10%. The most frequent cause of variability was incomplete segmentation due to an irregular shape or irregular margins.
Conclusion: In a minority (approximately 11%) of small solid intraparenchymal nodules, semiautomated measurements are not completely reproducible and, thus, may cause errors in the assessment of nodule growth. For small or irregularly shaped nodules, an observer should check the segmentation shown by the program.
(c) RSNA, 2006.