Estimation of the Probability of Malignancy in Solitary Pulmonary Nodules: Comparative Study of Conventional Interpretation Method and Bayesian Analysis.
10.3348/jkrs.1998.38.1.67
- Author:
Yu Whang OH
1
;
Seung Yong PARK
;
Eun Young KANG
;
Jai Soung PARK
;
Ki Yeol LEE
;
Hong In KIM
;
Young Nam KIM
;
Won Hyuck SUH
Author Information
1. Department of diagnostic Radiology, Korea University College of Medicine.
- Publication Type:Comparative Study ; Original Article
- Keywords:
Lung neoplasms, diagnosis;
Radigraphy, comparative studies;
Computed tomography(CT);
Lung, nodule
- MeSH:
Bayes Theorem*;
Humans;
Radiography, Thoracic;
ROC Curve;
Smoke;
Smoking;
Solitary Pulmonary Nodule*;
Tomography, X-Ray Computed
- From:Journal of the Korean Radiological Society
1998;38(1):67-74
- CountryRepublic of Korea
- Language:Korean
-
Abstract:
PURPOSE: The aim of this study was to compare the accuracy of the conventional method and Bayesian analysisin estimating the probability of malignancy in solitary pulmonary nodules. MATERIALS AND METHODS: The studyinvolved 83 pathologically proven cases of solitary pulmon ary nodules, 44 of which were malignant, and 39,benign. To estimate the probability of malignancy in solitary pulmonary nodules, chest radiographs and CT scans ofthe 83 patients were interpreted by a team of six ; three of the six used the conventional method of subjectiveinterpretation and the others. Bayesian analysis. The smoking history of 59 of the patients was obtained, and itwas decided whether this would help determine the probability of malignancy. RESULTS: On average, those using theconventional method correctly interpreted 34.7(78.9%) of 44 cases of malignant nodules and 27.7(71%) of 39 benignnodules, while those using Bayesian analysis correctly classified 32.3 cases of malignant nodules(73.4%) and 25.3cases of benign nodules(64.9%). Between the two teams, there was no statistically significant difference in theaccuracy of qualitative assessment(P>.05). In ROC analysis conventional interpretation and Bayesian analysisshowed an accuracy of Az=80.8 and Az=76.7, respectively. Among 59 patients known to have smoked, the conventionalmethod showed an accuracy of Az=79.0 without this knowledge and Az=80.2 with the knowledge for Bayesian analysis,the corresponding figures were Az=77.2 and Az=72.5, respectively. Information relating to smoking history thus didnot significantly improve the accuracy of prediction(p>.05). CONCLUSION: For estimating the probability ofmalignancy in solitary pulmonary nodules, the accuacy of the conventional method of interpretation is notsignificantly different from that of Bayesian analysis; information relating to smoking history significantlyimprove the accuracy of neither method.
