Click here to


Are you sure ?

Yes, do it No, cancel

Molecular Subtype Classification in Low Grade Glioma with Accelerated Diffusion Tensor Imaging

E Aliotta1*, H Nourzadeh1 , P Batchala2 , S Mukherjee2 , S Patel2 , (1) University of Virginia Department of Radiation Oncology, Charlottesville, VA, (2) University of Virginia Department of Radiology, Charlottesville, VA


(Tuesday, 7/16/2019) 4:30 PM - 6:00 PM

Room: 304ABC

Purpose: To classify diffuse lower grade gliomas (LGG) by molecular subtype using accelerated diffusion tensor imaging (DTI) scans.

Methods: LGG patients (N=36) (WHO grade II/III) with known isocitrate dehydrogenase (IDH) mutation and 1p/19q codeletion status were imaged preoperatively with DTI (b=1000s/mm2 along 20 directions plus one b=0 image). Whole lesion volumes were autodelineated using conventional sequences (T1 and T2 weighted) from contrast-enhanced MRI using the DeepMedic algorithm. DTI was reconstructed using linear-least-squares to generate apparent diffusion coefficient (ADC) and conventional fractional anisotropy (FAconv) maps. FA maps were also estimated from three-direction subsets of the DTI scans using previously described DIffNet method (FADiffNet), a pre-trained neural network that directly computes FA from raw, undersampled DTI data at each pixel. Mean whole-tumor ADC, FAconv, and FADiffNet were used to classify tumors according to IDH mutation status (IDH-wild type vs. IDH-mutant) and 1p/19q codeletion status (codeleted vs. noncodeleted). Inter-group differences were assessed using t-tests and receiver operator curve (ROC) analysis. Combined discriminatory performance of ADC+FA was assessed using binomial logistic regression.

Results: For IDH-wild type versus IDH-mutant LGGs, significant differences were observed in ADC (1.17±0.15 vs. 1.36±0.23 x10-3mm2/s, p=0.005), FAconv (0.23±0.05 vs. 0.17±0.04, p<0.001), and FADiffNet (0.20±0.05 vs. 0.14±0.03, p<0.001). FADiffNet outperformed FAconv and ADC (area under the curve, AUC=0.90 vs. 0.83 and 0.88). Combined ADC+FAconv and ADC+FADiffNet both slightly improved classification (AUC=0.91). Amongst IDH-mutant LGGs, 1p/19q codeleted versus noncodeleted groups showed significant differences in ADC (1.22±0.13 vs. 1.47±0.23 x10-3mm2/s, p=0.004) and non-significant differences in FAconv (0.19±0.04 vs. 0.16±0.03, p=0.17) and FADiffNet (0.15±0.04 vs. 0.13±0.01, p=0.09). FADiffNet did not classify 1p/19q codeletion as well as FAconv (AUC=0.63 vs. 0.67). ADC+FAconv slightly improved on ADC alone (AUC=0.78 vs. 0.77).

Conclusion: FA estimates made from 3-direction DTI scans (i.e. DWI) add value in classifying IDH mutation status and may aid 1p/19q codeletion status classification.


Brain, Diffusion, MRI


IM- MRI : Diffusion MRI

Contact Email