Differentiating the malignancy grades of thymic epithelial tumors using textural features of tumor heterogeneity on FDG PET/CT

Objectives Texture analysis of various kinds of cancer which gives information about intratumoral heterogeneity is increasingly conducted using FDG PET. We aimed to explore the ability of textural features of thymic epithelial tumors (TETs) on FDG PET/CT in differentiating their grade of malignancy.

Methods We retrospectively studied 140 patients with pathologically proven TETs (76 male and 64 female; mean age, 54±12 years) who underwent FDG PET/CT. TETs were classified according to the pathology results into three grades with increasing order of malignancy, i.e. low-risk thymoma (LRT; WHO classification A, AB and B1), high-risk thymoma (HRT; B2 and B3), and thymic carcinoma (TC). We examined the ability of PET parameters including SUVmax, and local (busyness, coarseness, complexity, contrast, and entropy) and regional textural features ((High/Low Gray-level Short/Long) Run-length Emphases) calculated within CT-based volumes of interest for discriminating the malignancy grades of TETs. The ROC curve was used in univariate analysis, and the multinomial cumulative logistic regression was used in multivariate analysis to demonstrate the independent significance of covariates.

Results In differentiating LRT from HRT and TC, fair or higher accuracies (area under the ROC curve (AUC) > 0.7) were achieved by SUVmax (AUC = 0.860), complexity (0.780), LGSRE (0.746) and LRE (0.728), SRE (0.736) (all P < 0.0001). In differentiating TC from LRT and HRT, fair or higher accuracies were achieved by SUVmax (0.905), complexity (0.736), HGLRE (0.728), LGRE (0.743), LGSRE (0.793), and SRE (0.719) (all P < 0.0001). In multivariate analysis, complexity (P = 0.007), busyness (P = 0.014), LRE (P = 0.001) and SRE (P < 0.001) remained significant independently of SUVmax.

Conclusions Texture analysis of FDG PET has a potential in differentiating the malignancy grades of TETs. Subsequent studies should focus on deciding optimal algorithms to calculate textural features that best correlate with the malignancy grades of TETs.