AUTHOR=Deng Da-Biao , Liao Yu-Ting , Zhou Jiang-Fen , Cheng Li-Na , He Peng , Wu Sheng-Nan , Wang Wen-Sheng , Zhou Quan TITLE=Non-Invasive Prediction of Survival Time of Midline Glioma Patients Using Machine Learning on Multiparametric MRI Radiomics Features JOURNAL=Frontiers in Neurology VOLUME=Volume 13 - 2022 YEAR=2022 URL=https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2022.866274 DOI=10.3389/fneur.2022.866274 ISSN=1664-2295 ABSTRACT=Objectives: To explore the feasibility of predicting overall survival (OS) of midline gliomas using multiparameter MRI features. Methods: Eighty-four patients with midline gliomas were retrospectively collected, including 40 patients with OS>12 months and 44 patients with OS<12 months. The features were extracted from the largest slice of the tumor which was manually segmented on the T2-weighted (T2w), fluid attenuated inversion recovery (FLAIR) and contrast-enhanced T1 weighted (T1c) images. The data were randomly divided into 70% training cohort and 30% test cohort, and were normalized and standardized by Z-score. Feature dimensionality reduction was performed by using variance method and Max-Relevance and Min-Redundancy (mRMR). Using logistic regression algorithm to construct three models of T2w, FLAIR, T1c and one combined model. The test cohort was used to evaluate the models, and receiver operating characteristic (ROC), area under curve (AUC), sensitivity, specificity, and accuracy were calculated. The nomogram of the combined model was built and evaluated by calibration curve. Decision curve analysis (DCA) was used to evaluate the clinical application value of the four models. Results: A total of 1316 features were extracted from T2w, FLAIR and T1c images, respectively. Four models were established by using optimal features. In the test cohort, the performance of combined model was the best compared to other models. The area under curve (AUC) of T2w, FLAIR, T1c and combined models was 0.73, 0.78, 0.74, 0.87, while the accuracy was 0.72, 0.76, 0.72, 0.84, respectively. The ROC curve and DCA showed that the combined model had better efficiency and more favorable clinical benefits. Conclusion: The combined radiomics model based on multiparameter MRI can provide a reliable noninvasive method for prognostic prediction of midline gliomas.