Click here to


Are you sure ?

Yes, do it No, cancel

Machine-Learning-Based Classification of Glioblastoma Using Dynamic Susceptibility Enhanced MR Image Derived Delta-Radiomic Features

J Jeong*, L Wang , J Bing , Y Lei , T Liu , A Ali , W Curran , H Mao , X Yang , Emory University, Atlanta, GA


(Wednesday, 8/1/2018) 7:30 AM - 9:30 AM

Room: Room 202

Purpose: Glioblastoma (GBM) is the most aggressive glioma with poor prognosis due to its heterogeneity. The purpose of this study is to improve the tissue characterization of these highly heterogeneous tumors using delta-radiomic features of dynamic susceptibility enhanced (DSE) MR images, which are commonly used to derive blood perfusion to the tumor, with machine learning approaches.

Methods: Multiparametric magnetic resonance (MR) images of 18 patients with GBM, with pathologically confirmed 11 high and 7 low grade, were taken. All DSE images and tumors were registered to and contoured in FLAIR images. These contours and its mirror images were used to extract delta-radiomic features in the DSE volumes before applying feature selection methods. The most salient features were selected to train a random forest to differentiate high-grade (HG) and low-grade (LG) GBMs while feature correlation limits were applied to remove redundancies. Then a leave-two-out, one HG and LG GBM, cross-validation random forest was applied to the dataset to classify GBMs. To evaluate the performance of random forest predictions, overall prediction accuracy, confidence, sensitivity and specificity were calculated.

Results: Analysis of the predictions showed that our method consistently predicted 16 out of 18 patients correctly (0.89). Based on the leave-two-out cross-validation, mean prediction accuracy was 0.92±0.1 for HG and 0.70±0.31 for LG. The area under the receiver operating characteristic curve was 0.85.

Conclusion: Our method performed well in classifying high and low grade GBMs based on the DSE MRI data. This study shows that delta-radiomic features of DSE MRI are correlated with GBM disease grades that may elucidate the underlying tumor biology and response to therapy. The performance of our method in characterizing DSE MRIs of GBMs will be explored further using temporal delta-radiomic features that take advantage of the differences in tumor contrast between the baseline and peak contrast images.


Not Applicable / None Entered.


IM/TH- Image Analysis (Single modality or Multi-modality): Computer-aided decision support systems (detection, diagnosis, risk prediction, staging, treatment response assessment/monitoring, prognosis prediction)

Contact Email