Application Of Multimodal Magnetic Resonance Imaging And Radiomics In Evaluation Of Molecular Profiles And Biological Function Of Glioma

Part Ⅰ:Application of MRI-based cerebral oxygen metabolism imaging and dynamic contrast-enhanced imaging in predicting genetic profile and correlation with blood-oxygen biological functionObjective:To explore the predictability of MRI-based QSM+q BOLD oxygen metabolism imaging using cluster analysis of time evolution（CAT）,in combination with DCE-MRI in prediction of genetic profiling,and correlation with blood-oxygen biological function.Methods:91 patients with histopathologically confirmed glioma were examined with MRI-based cerebral oxygen metabolism mapping and DCE-MRI.Imaging biomarkers,including oxygen metabolism（OEF）and angiogenesis(volume transfer constant^[Ktrans],cerebral blood volume[CBV],and cerebral blood flow[CBF]),were investigated to predict IDH1mutation,MGMT promoter methylation status,receptor tyrosine kinase（RTK）subgroup,and differentiation of glioblastoma（GBM）versus lower-grade glioma（Ler GG）.Results were compared with DCE-MRI using ROC analysis.KEGG enrichment analysis was used to explore the correlation between biological function and imaging phenotypes.Results:IDH1-mutated lower-grade gliomas exhibited significantly lower OEF and hypoperfusion than IDH1 wild-type tumors（all P<0.01）.OEF and perfusion metrics showed a tendency toward higher values in MGMT unmethylated glioblastoma but only OEF retained significance（P=0.01）.Relative prevalence of RTK alterations were associated with increased OEF（P=.003）and perfusion values（P<.05）.ROC analysis suggested OEF achieved best performance for IDH1 mutation detection（AUC=0.828）.None of the investigated parameters enabled prediction of MGMT status except OEF with a moderate AUC of 0.784.Predictive value for RTK subgroup was acceptable by using OEF（AUC=0.764）and CBV（AUC=0.754）.OEF and perfusion metrics demonstrated excellent performance in gliomas grading,parameters with AUC>0.8 were OEF（AUC=0.810）,K^trans（AUC=0.819）,and CBV（AUC=0.802）,respectively.Moreover,mutational land and KEGG analysis revealed hypoxia or angiogenesis-relevant gene signatures were associated with specific imaging phenotypes.Conclusion:Cluster analysis of time evolution for MRI-based QSM+q BOLD model is a promising technology for oxygen measurement,and along with perfusion MRI can predict genetic profiles and tumor hypoxia/angiogenesis in a noninvasive and clinically relevant manner.Part Ⅱ:Non-invasive Evaluation of Notch Signaling Mutation via Radiomic Signatures Based on Multiparametric MRI in Association with Biological Functions of Patients with Glioma: A Multi-Institutional StudyObjectives: To predict mutations in Notch signaling pathway in patients with glioma by using multiparametric MRI（mp MRI）radiomics and to explore biofunction correlation with downstream targets.Methods: The radiomics model was developed by using 48 retrospectively-collected patients with glioma who underwent next-generation DNA sequencing.For model performance,a retrospective analysis of 47 patients pathologically confirmed gliomas from two public databases（TCGA and CPTAC）were evaluated.Then,radiomics features were extracted from CE-T1 WI,T1WI,T2 WI,and T2 FLAIR and imaging signatures were selected using a LASSO regression algorithm.Diagnostic performance was compared using single imaging radiomics models and the combined mp MRI model.Incorporating clinical factors,a radiomic-clinical nomogram was constructed.The predictive performance（discrimination,calibration,and clinical usefulness）was assessed and validated in the test set.Finally,the radiomic signatures were validated by immunohistochemistry analysis.Results: The radiomics signature derived from the combination of all MRI modalities showed highest AUC for prediction of Notch signaling mutation in both sets（AUC,0.857 and 0.823,respectively）.The radiomics nomogram that incorporated radiomics signature and KPS status performed favorable calibration and discrimination in both sets,with AUCs of 0.891（95% CI,0.79-0.99）and 0.859（95% CI,0.75-0.97）.Decision curve analysis confirmed its clinical usefulness.Notch1 receptor and its downstream targets,Hes1,displayed reduced expression with an elevated expression of Dll4 and ASCL1,showing a general trend of increasing cancer characteristics,such as increased tumor cell density and angiogenesis.Conclusion: The proposed multiparametric MRI-based radiomics nomogram may reflect the intratumour heterogeneity associated with downstream biofunction that predict Notch signaling mutation in a noninvasive and clinically relevant manner.Part Ⅲ:Multiparametric Diffusion Magnetic Resonance Imaging and Radiomics in Prediction of IDH1 genotype of lower-grade gliomaObjectives: To evaluate the diagnostic value of multi-parameter diffusion magnetic resonance imaging combined with radiomics in the prediction of IDH1 genotype of lowergrade gliomas.Methods: Sixty-five patients with histopathologically proven lower-grade gliomas（Ler GG,WHO grade II/III）were prospectively collected and examined by diffusion magnetic resonance imaging（d MRI）.The multi-parameter diffusion MRI sequence was used a spinecho echo-planar imaging consisting of 25 non-collinear directions at three b values of 0,1250s/mm2 and 2500s/mm2.After eddy-currents and motion artifacts correction,bias field correction and brain extraction,diffusion tensor imaging（DTI）,free water imaging（FWI）and free water corrected DTI（FW_DTI）,neurite directed dispersion and density imaging（NODDI）,apparent fiber density imaging（AFD）,and generalized fractional anisotropy based on constant solid angle（CSA-GFA）were generated.The histogram features of the solid tumor and edema area are extracted from the parametric imaging and were reduced to the first 3 features in each diffusion model by classic minimum redundancy maximum relevance method.The principal component multivariate logistic regression analysis was used to compare the predictive efficacy of different diffusion models for the IDH1 genotype of lower-grade gliomas.Results: A total of 65 patients（31 males and 34 females;The mean age was 43 ± 12 years）were included in this study,including 35 WHO grade II gliomas（29 with IDH1 mutations）and 30 WHO grade III gliomas（17 with IDH1 mutations）.All single diffusion models can be used to predict the IDH1 genotype of lower-grade gliomas,with NODDI and AFD diffusion models achieved the best predictive performance（AUC = 0.801 and 0.807,respectively;P <0.01）。The diffusion model combined maximum apparent fiber density and free water-corrected FA histogram features had the highest prediction efficiency,with the area under the curve of 0.855（P <0.0001）.Conclusion: Multiparametric diffusion magnetic resonance imaging combined with radiomic analysis is a promising method for the prediction of IDH1 genotype of lower-grade gliomas,and the diffusion model combined maximum apparent fiber density and free watercorrected FA histogram features has the highest prediction efficiency.