A Comparative Study of Radiomic and Connectomic Approaches to Classification of IDH1 Status and 1p/19q Co-deletion in Lower Grade Gliomas

Purpose: Grade III and IV brain tumors are labeled high grade, or malignant. Lower grade tumors (grade II and III) can progress to high grade and must be closely monitored. In lower grade gliomas, the presence of a specific IDH1 gene mutation and the 1p/19q chromosomal co-deletion confer favorable prognosis and alternative treatment strategy. Presently, these markers are evaluated using surgically obtained tissue specimens. In this study, we evaluate noninvasive approaches to classification of these genetic markers. We hypothesized that connectomic and radiomic approaches to classification would perform similarly. We also tested combined classification, incorporating radiomics and connectomics. Methods: Binary classifiers used radiomic and connectomic features from MRI to classify IDH1 and 1p/19q co-deletion status. Radiomic features were calculated to characterize tumor gray-level, texture, and shape. Voxel-based morphometry was performed to create gray-matter structural connectomes. Nodal efficiencies of brain regions, number of nodes and connections were computed. Binary classifiers predicted IDH1 and 1p/19q co-deletion status. Statistical analysis quantified differences in model performance. Results: Connectomic and radiomic features had insignificant difference in classification of IDH1 status. Radiomic and connectomic classification of 1p/19q co-deletion status had no significant accuracy difference, however, radiomics had significantly higher AUC score. The combined approach had no significant difference to radiomics and connectomics except for a significantly higher AUC score than connectomics in 1p/19q co-deletion classification. Conclusion: Altogether, the study shows that radiomics, connectomics, and a combination of the two are viable classification approaches for these markers. Future studies could incorporate these methods to improve diagnostic performance. Keywords: Connectomics, Glioma, IDH, Machine Learning, Radiomics