Gliomas staging and prognosis has been recently related to the mutation of the gene encoding isocitrate dehydrogenase (IDH). Dynamic Susceptibility Contrast Perfusion-Weighted Imaging (DSC-PWI) is an established MRI technique for glioma staging and prognosis but its value for non-invasive IDH phenotyping has not been thoroughly investigated. Combining DSC-PWI with advanced image analysis methods, such as texture analysis and machine learning, has the potential to increase the diagnostic accuracy of DSC MRI for IDH mutation status detection.
Our retrospective, multi-centre study included 365 patients [184 female; 181 male, median age: 49 (range 21-81 years)], who had been immunohistopathologically diagnosed with gliomas (198 IDH positive; 167 IDH negative). A fully adaptive Bayesian method was applied to calculate leakage-corrected relative cerebral blood volume (rCBV) maps from the DSC raw data. Tumour boundaries were manually defined and co-registered to the rCBV maps. The texture features were calculated based on the rCBV findings within the defined tumour. A 2-fold cross-validation setting of 1000 iterations using support vector machine and multinomial ordinal regression was applied to assess the predictive power of the extracted features in IDH phenotyping.
The proposed rCBV analysis for IDH status stratification showed a sensitivity rate of 75% and specificity rate of 88%. In the non-parametric Wilcoxon test nine out of the ten classical histogram statistics and 12 texture features appeared significantly different across mutation status (p < 0.02). The distance error (difference between the real and predicted grade) was used to show the classification across grading: In 90.2 % of the cases the same features led to inferior or equal to 1 in and in 71.3% of cases to an exact prediction.
These promising preliminary results for DSC-MRI-based glioma stratification with respect to their IDH mutation status suggest further exploration into the potential diagnostic and predictive and predictive value of rCBV analysis as surrogate non-invasive biomarker for IDH classification of gliomas.
Clinical trial identification
Legal entity responsible for the study
University College London.
Has not received any funding.
D. Roettger: Head of Scientific and Medical Affairs: IAG. All other authors have declared no conflicts of interest.