Comparison of Radiomics Analyses Based on Different Magnetic Resonance Imaging Sequences in Grading and Molecular Genomic Typing of Glioma

• Published in: Journal of computer assisted tomography Vol. 45; no. 1; p. 110
• Main Authors: Huang, Wei-Yuan, Wen, Ling-Hua, Wu, Gang, Hu, Ming-Zheng, Zhang, Chao-Cai, Chen, Feng, Zhao, Jian-Nong
• Format: Journal Article
• Language: English
• Published: United States 01.01.2021
• Subjects: Adolescent; Adult; Aged; Brain Neoplasms - diagnostic imaging; Brain Neoplasms - genetics; Brain Neoplasms - pathology; Child; DNA Methylation; DNA Modification Methylases - genetics; DNA Repair Enzymes - genetics; Female; Glioma - diagnostic imaging; Glioma - genetics; Glioma - pathology; Humans; Isocitrate Dehydrogenase - genetics; Logistic Models; Magnetic Resonance Imaging - methods; Male; Middle Aged; Mutation; Neoplasm Staging; Promoter Regions, Genetic; Radiographic Image Interpretation, Computer-Assisted - methods; Tumor Suppressor Proteins - genetics; Young Adult
• ISSN: 1532-3145
• DOI: 10.1097/RCT.0000000000001114

To investigate the value of radiomics analyses based on different magnetic resonance (MR) sequences in the noninvasive evaluation of glioma characteristics for the differentiation of low-grade glioma versus high-grade glioma, isocitrate dehydrogenase (IDH)1 mutation versus IDH1 wild-type, and mutation status and 6-methylguanine-DNA methyltransferase (MGMT) promoter methylation (+) versus MGMT promoter methylation (-) glioma. Fifty-nine patients with untreated glioma who underwent a standard 3T-MR tumor protocol were included in the study. A total of 396 radiomics features were extracted from the MR images, with the manually delineated tumor as the volume of interest. Clinical imaging diagnostic features (tumor location, necrosis/cyst change, crossing midline, and the degree of enhancement or peritumoral edema) were analyzed by univariate logistic regression to select independent clinical factors. Radiomics and combined clinical-radiomics models were established for grading and molecular genomic typing of glioma by multiple logistic regression and cross-validation. The performance of the models based on different sequences was evaluated by using receiver operating characteristic curves, nomograms, and decision curves. The radiomics model based on T1-CE performed better than models based on other sequences in predicting the tumor grade and the IDH1 status of the glioma. The radiomics model based on T2 performed better than models based on other sequences in predicting the MGMT methylation status of glioma. Only the T1 combined clinical-radiomics model showed improved prediction performance in predicting tumor grade and the IDH1 status. The results demonstrate that state-of-the-art radiomics analysis methods based on multiparametric MR image data and radiomics features can significantly contribute to pretreatment glioma grading and molecular subtype classification.