Background and purpose: Pulmonary squamous cell carcinoma (SCC) was reported to have significantly higher histogram_kurtosis obtained by unenhanced computed tomography (CT) and a greater maximum standardized uptake value (SUVmax) determined by 18F-fluorodeoxyglucose positron emission tomography/computed tomography (18F-FDG-PET/CT) than pulmonary adenocarcinoma (AD). The first study aim was to investigate whether CT radiomics features could differentiate SCC from AD. The second aim was to investigate correlations between CT radiomics features and SUVmax.
Materials and methods: Forty solid lung tumors < 3 cm in diameter pathologically proven to be SCC (n = 18) or AD (n = 22) were included. The SUVmax was determined by 18F-FDG-PET/CT and 42 CT radiomics features were obtained from unenhanced CT. A Mann-Whitney U test was used to compare the SUVmax and each CT radiomics feature to differentiate SCC from AD. The area under the curve (AUC) was calculated from receiver operating characteristic (ROC) analysis. A binary logistic regression analysis was performed for a combination of SUVmax with each CT radiomics feature. A Spearman rank correlation analysis was performed to determine correlations between SUVmax and CT radiomics features.
Results: Twenty-six of 42 CT radiomics features were significantly different between groups. The best AUC of the CT radiomics features was 0.81 for both the GLRLM_SRHGE and GLZLM_HGZE. The AUC value improved when the shape_sphericity feature was combined with SUVmax (AUC = 0.92). The CT radiomics features exhibiting strongest correlations with the SUVmax were different in both AD and SCC groups.
Conclusion: CT radiomics features correlated with tumor structural characteristics of SCC and AD, and were closely related to metabolic information from FDG-PET/CT.
Keywords: Computed tomography; GLCM; Lung cancer; Positron emission tomography; Radiomics; Texture analysis.
Copyright © 2020 Elsevier B.V. All rights reserved.