CT Radiomic Features for Predicting Resectability and TNM Staging in Thymic Epithelial Tumors

Abstract

Background: To explore the performance of a computed tomography based radiomics model in the preoperative prediction of resectability status and TNM staging in thymic epithelial tumors.

Methods: We reviewed the last preoperative computed tomography scan of patients with thymic epithelial tumors prior to resection and pathology evaluation at our institution between February 2008 and June 2019. A total of 101 quantitative features were extracted and a radiomics model was trained using elastic net penalized logistic regressions for each aim. In the set-aside testing sets, discriminating performance of each model was assessed with area under receiver operating characteristic curve.

Results: Our final population consisted of 243 patients with: 153 (87%) thymomas, 23 (9%) thymic carcinomas, and 9 (4%) thymic carcinoids. Incomplete resections (R1 or R2) occurred in 38 (16%) patients, and 67 (28%) patients had more advanced stage tumors (stage III or IV). In the set-aside testing sets, the radiomics model achieved good performance in preoperatively predicting incomplete resections (area under receiver operating characteristic curve: 0.80) and advanced stage tumors (area under receiver operating characteristic curve: 0.70).

Conclusions: Our computed tomography radiomics model achieved good performance to predict resectability status and staging in thymic epithelial tumors, suggesting a potential value for the evaluation of radiomic features in the preoperative prediction of surgical outcomes in thymic malignancies.

Figure 1:. Flowchart of patient selection.

Patients with free surgical margins (R0) were included in the resected group, while patients with microscopic (R1) or macroscopic (R2) residual tumor were included in the unresected group. Early-stage thymic epithelial tumors (TETs) include TNM stages I and II, and advanced-stage TETs include TNM stages III and IV.

Figure 2:. Preoperative chest computed tomography of a 53-year-old female patient presenting with a mediastinal mass

(a). Tumor was manually segmented (b) and radiomic features (c) were extracted. Note the tumor heterogeneity throughout the lesion illustrated in an entropy map in (c), where the blue areas represent tumor areas with low entropy. Final diagnosis was an advanced-stage thymoma (Masaoka-Koga IV) with positive margins due to microscopic residual tumor (R1).

Figure 3:. Preoperative chest computed tomography of a 55-year-old female patient presenting with a mediastinal mass

(a). After tumor segmentation (b), radiomic features were extracted (c). Note that the lesion has a relative homogenous appearance (uniformly orange/red – high entropy) in the entropy map (c). The lesion (early-stage thymoma – Masaoka-Koga II) was completely resected with no evidence of residual disease (R0).

Figure 4:. Correlation matrix heat map of selected radiomics features.

The magnitude of the correlation is indicated in the color bar.

Figure 5:. Model performance for the prediction of resectability in thymic epithelial tumors.

Best performing features listed by importance (a) and receiver operating characteristic curve for the predictive model (b). Of note: RE_RLM: reflects runs of similar intensity in the region of interest (ROI) and a lower value indicates a more homogeneous appearing ROI; JOINT_ENTROPY_GLCM: measures the randomness of the ROI appearance; VAR_FO: reflects the spread in the data and a higher variance indicates a higher spread of Hounsfield units in the ROI; SUM_VAR_GLCM: assess dispersion in the ROI (the more spread out the intensities across the image, the higher the value).

Figure 6:. Model performance for the prediction of TNM stage in thymic epithelial tumors.

Best performing features listed by importance (a) and ROC curve for the predictive model (b). Of note: LRHGLE_RLM: this metric looks at long runs of similar high gray levels in the image; LZLGLE_SZM: this parameter focuses on large areas of similar low gray levels; DIFF_VAR_GLCM: this feature is a measure of heterogeneity that places greater emphasis on strongly contrasting pixel pairs i.e., an emphasis on a very heterogenous appearance; CLUST_PROMIN_GLCM: this metric is related to the asymmetry in the data.