Accuracy of high-resolution CT in the diagnosis of diffuse lung disease: effect of predominance and distribution of findings.

Objective: The purpose of this study was to determine whether the predominant findings at high-resolution CT influence the accuracy of diagnosis of diffuse lung disease.

Materials And Methods: The cases of 100 patients with diffuse lung disease who underwent high-resolution CT and tissue diagnosis were studied. Three thoracic radiologists reviewed high-resolution CT images blindly and independently for patterns of abnormality, listing their three main diagnoses and level of confidence in the first choice. The effect of the findings on accuracy was analyzed.

Results: For honeycombing, the accuracy of the main diagnosis was 96.6%, 92.2%, and 92.3% for the three readers, and that of the three main diagnoses was 96.6%, 96.1%, and 92.3%. For cysts, the accuracy of the main diagnosis was 88.9%, 80%, and 81.8% and of the three main diagnoses was 100%, 90%, and 90.9%. For bronchovascular thickening, the accuracy of the main diagnosis was 91.7%, 87.5%, and 90.9% and of the three main diagnoses was 91.7%, 100%, and 90.9%. For ground-glass opacification (GGO), the accuracy of the main diagnosis was 75.5%, 55%, and 44.2% and of the three main diagnoses was 89.8%, 75%, and 65.4%. Only combining honeycombing with GGO improved the accuracy of GGO. Anatomic craniocaudal distribution improved reader accuracy when GGO was predominantly present in the lower part of the lung. Interobserver agreement on the presence of major findings was a mean kappa value of 0.45 for honeycombing, 0.74 for lung cysts, 0.63 for bronchovascular thickening, and 0.56 for GGO. Agreement for the craniocaudal distribution of major findings was a mean kappa value of 0.48 for honeycombing, 0.52 for bronchovascular thickening, and 0.32 for GGO.

Conclusion: The predominant findings of honeycombing and bronchovascular thickening are associated with more than 90% accuracy in the first-choice diagnosis of diffuse lung disease; the finding of lung cysts has 80-89% accuracy. GGO as a predominant pattern had unreliable accuracy, but the accuracy improved when GGO was combined with either honeycombing or lower-lung distribution.