A complete pipeline for glioma grading using intelligible AI on multimodal MRI data

1) Objectives: Machine learning for binary glioma grading have been extensively used on anatomical MRI, especially using the BraTS dataset. The relevance of radiomic criteria based multimodal imaging, including diffusion, perfusion and spectroscopy data is to be explored, as datasets are scarce, there no common benchmark performance comparison. 2) Material methods: Poitiers University Hospital provides 123 patient data. We computed 124 features let a recursive feature elimination algorithm (RFE) yield relevant, reduced subset features. trained SVM classifier this subset. proposed method adapt dataset allow comparison with literature. got reference point by training only, showed improvements when multimodalities were added. explored through RFE not constant induce variability in performances. To smooth variability, we applied 100 times incremented selected features, resulting global ranking. also show best reached these trainings its 3) Results: 86.5% accuracy, mean accuracy 78.6%. rankings shows that sequences most relevant grading, T1 post-gadolinium, cerebral blood volume flow. Intensity texture frequently selected, while anisotropic diffusion coefficient, time peak transit mappings seem irrelevant. 4) Conclusion: Multimodal radiomics improve classification consistent radiological analysis.