Quantitative 3D Shape Analysis of CT Images of Thymoma: A Comparison With Histological Types.

Abstract

OBJECTIVE. The purpose of this study is to differentiate between low- and high-risk types of thymoma using quantitative 3D shape analysis of CT images. MATERIALS AND METHODS. This retrospective study included 44 patients with a pathologic diagnosis of thymoma. Two radiologists semiautomatically contoured CT images of the tumors and evaluated 3D shape parameters-namely, quantitative indicators of surface smoothness, including sphericity, ellipsoidality, and discrete compactness. The visual CT findings that were analyzed included longest diameter, shape (round-oval, lobulated, or irregular), calcification, cystic or necrotic changes, and enhancement pattern (homogeneous or heterogeneous). The difference and discriminating performance between low-risk (types A, AB, and B1) and high-risk (types B2 and B3) thymomas were statistically assessed. Interobserver agreement was determined using the concordance correlation coefficient. RESULTS. Twenty-three low-risk and 21 high-risk thymomas were identified on the basis of pathologic findings. The median values of sphericity and ellipsoidality were significantly higher for low-risk thymomas than for high-risk thymomas (for sphericity, 0.566 vs 0.517; for ellipsoidality, 0.941 vs 0.875; p < 0.05 for both). The AUC values of sphericity and ellipsoidality were 0.704 and 0.712, respectively. The best cutoff values were 0.528 and 0.919 for sphericity and ellipsoidality, respectively. Risk assessment combining these cutoff values and the mode of tumor detection (incidental detection or detection based on the presence of symptoms) improved the AUC value to 0.856 (sensitivity, 81.0% [17 of 21 patients]; specificity, 82.6% [19 of 23 patients]). All 3D shape parameters showed almost perfect interobserver agreement (concordance correlation coefficient, > 0.90). The visual CT findings were not significantly different between low- and high-risk thymomas (p > 0.05 for all). CONCLUSION. Quantitative 3D shape analysis has excellent reproducibility, and combining this technique with information on the detection mode helps differentiate low- from high-risk thymomas.