High-order radiomics features based on T2 FLAIR MRI predict multiple glioma immunohistochemical features: A more precise and personalized gliomas management

• DOI: 10.1371/journal.pone.0227703
• Publication Date: 2020-01-22T18:33:47Z
• Authors: Jing Li; Siyun Liu; Ying Qin; Yan Zhang; Ning Wang; Huaijun Liu
• Keywords: Fluid-attenuated inversion recovery; Lasso

Objective To investigate the performance of high-order radiomics features and models based on T2-weighted fluid-attenuated inversion recovery (T2 FLAIR) in predicting immunohistochemical biomarkers glioma, order to execute a non-invasive, more precise personalized glioma disease management. Methods 51 pathologically confirmed gliomas patients committed our hospital from March 2015 June 2018 were retrospective analysis, Ki-67, vimentin, S-100 CD34 data collected. The volumes interest (VOIs) manually sketched extracted. Feature reduction was performed by ANOVA+ Mann-Whiney, spearman correlation least absolute shrinkage selection operator (LASSO) Gradient descent algorithm (GBDT). SMOTE technique used solve bias between two groups. Comprehensive binary logistic regression established. Area under ROC curves (AUC), sensitivity, specificity accuracy evaluate predict models. Models reliability decided according standard net benefit decision curves. Results Four clusters significant screened out four constructed. AUC S-100, vimentin 0.713, 0.923, 0.854 0.745, respectively. sensitivities 0.692, 0.893, 0.875 0.556, specificities were: 0.667, 0.905, 0.722, 0.875, with 0.660, 0.898, 0.738, According curves, had reference values. Conclusion T2 FLAIR can potentially expression. Radiomics model expected be computer-intelligent, accurate management method for gliomas.