Machine learning assisted DSC-MRI radiomics as a tool for glioma classification by grade and mutation status
AbstractBackgroundMachine learning assisted MRI radiomics, which combines MRI techniques with machine learning methodology, is rapidly gaining attention as a promising method for staging of brain gliomas. This study assesses the diagnostic value of such framework applied to dynamic susceptibility contrast (DSC)-MRI in classifying treatment-naïve gliomas from a multi-center patient pool into WHO grades II-IV and across their isocitrate dehydrogenase (IDH) mutation status.Methods333 patients from 6 tertiary centres, diagnosed histologically and molecularly with primary gliomas (IDH-mutant=151 or IDH-wildtype=182) were retrospectively identified. Raw DSC-MRI data was post-processed for normalised leakage-corrected relative cerebral blood volume (rCBV) maps. Shape, intensity distribution (histogram) and rotational invariant Haralick texture features over the tumour mask were extracted. Differences in extracted features between IDH-wildtype and IDH-mutant gliomas and across three glioma grades were tested using the Wilcoxon two-sample test. A random forest algorithm was employed (2-fold cross-validation, 250 repeats) to predict grades or mutation status using the extracted features.ResultsFeatures from all types (shape, distribution, texture) showed significant differences across mutation status. WHO grade II-III differentiation was mostly driven by shape features while texture and intensity feature were more relevant for the III-IV separation. Increased number of features became significant when differentiating grades further apart from one another. Gliomas were correctly stratified by IDH mutation status in 71% of the cases and by grade in 53% of the cases. In addition, 87% of the gliomas grades predicted with an error distance up to 1.ConclusionDespite large heterogeneity in the multi-center dataset, machine learning assisted DSC-MRI radiomics hold potential to address the inherent variability and presents a promising approach for non-invasive glioma molecular subtyping and grading.Key points-On highly heterogenous, multi-centre data, machine learning on DSC-MRI features can correctly predict glioma IDH subtyping in 71% of cases and glioma grade II-IV in 53% of the cases (87% <1 grade difference)-Shape features distinguish best grade II from grade III gliomas.-Texture and distribution features distinguish best grade III from grade IV tumours.Importance of studyThis work illustrates the diagnostic value of combining machine learning and dynamic susceptibility contrast-enhanced MRI (DSC-MRI) radiomics in classifying gliomas into WHO grades II-IV as well as across their isocitrate dehydrogenase (IDH) mutation status. Despite the data heterogeneity inherent to the multi-centre design of the studied cohort (333 subjects, 6 centres) that greatly increases the theoretical challenges of machine learning frameworks, good classification performance (accuracy of 53% across grades (87% <1 grade difference) and 71% across mutation status) was obtained. Therefore, our results provide a proof-of-concept for this emerging precision medicine field that has good generalisability and scalability properties. Introspection on the classification errors highlighted mostly borderline cases and helped underline the challenges of a categorical classification in a pathological continuum.With its strong generalisability property, its ability to further incorporate participating centres and its possible use to identify borderline cases, the proposed machine learning framework has the potential to contribute to the clinical translation of machine-learning assisted diagnostic tools in neuro-oncology.
