Click here to


Are you sure ?

Yes, do it No, cancel

Quantitative Analysis of Chemical Substances in Brain Gliomas by MRI-DWI Radiomics

H Deng1*, G Gong1 , Q Qiu1 , Y Yin1 , (1) Shandong Cancer Hospital Affiliated to Shandong University, Shandong Academy of Medical Sciences


(Monday, 7/30/2018) 9:30 AM - 10:00 AM

Room: Exhibit Hall | Forum 8

Purpose: Choline/N-acetyl-aspartate (Cho/NAA) was the main reference to diagnose and detect the recurrence of the glioma. In this study, we quantified MRI- diffusion weighted imaging (MRI-DWI) radiomics features for the content of chemical substances of brain gliomas.

Methods: Magnetic resonance spectrum (MRS) images of 63 patients with glioma were analyzed retrospectively. ROI of DWI (ROID) was delineated on DWI images, avoiding large vessels, ventricles, fats and necrotic areas according the MRS reference images, and 1059 ROID were obtained. The Cho/Naa value of ROID was collected. 85 radiomic features were extracted using Radiomics analysis software, including 8 Gradient Orient Histogram (GHO) feature, 22 Gray Level Co-occurrence Matrix (GLCM) feature, 11 Gray Level Run Length Matrix (GLRLM) feature, 39 Intensity Histogram (IH) feature, and 5 Neighbor Intensity Difference Matrix (NGTDM) feature. The value of Cho/Naa was divided into five groups (0-0.5, 0.51-1, 1.01-1.5, 1.51-3, >3). The radiomic features were normalized by Z-score, and it were screened by using the Least absolute shrinkage and selection operator (LASSO) with ten-fold cross validation method. The relation of radiomic features and the value of Cho/Naa were analyzed.

Results: 56 texture parameters were reserved after removing the redundancies. We removed the parameters that no significant difference between groups in order to get a better prognostic model, and 45 texture parameters were reserved. The optimal prognostic model identified ten imaging biomarkers that quantified the content of chemical substances of tumor by using Automatic Linear Modeling. In the validation cohort, we used these 10 parameters to establish the prediction model of Cho/Naa by using the LASSO, and the prediction accuracy of our prognostic model reached 85.9%.

Conclusion: Radiomic features could predict the content of chemical substances in gliomas with high accuracy, which may become a novel method for diagnosis and detection of glioma recurrence.




Not Applicable / None Entered.

Contact Email