Preliminary Study On Prognosis And Genotype Of Glioma Based On MR Morphology And Radiomics

Part 1 Predicting survival of high-grade gliomas by early FLAIR hyperintense around residual cavity: A preliminary study of radiomicsObjective: Prognostic evaluation of high-grade gliomas(HGG) is difficult.The aim of this study was to develop a radiomic model based on FLAIR images and to evaluate this model in predicting long time survival states of patients with HGG after resection and chemoradiotherapy.Methods:This retrospective study enrolled consecutive 77 HGG patients(grade Ⅲ 34 cases and grade Ⅳ 43 cases) with FLAIR hyperintense in the surrounding area of residual cavity.All these patients underwent gross-total resection of tumors followed by concurrent conventional chemoradiotherapy and adjuvant chemotherapy.We segmented area with FLAIR hyperintense and extracted radiomic features.Then we built radiomic,clinical and combined models with the algorithms of random forest(RF)and logistic regression(LR).The predicting efficiency of three models with two different algorithms on favorable survival states were compared.Results:The overall survival(OS)and progression-free survival(PFS)of favorable prognosis group(OS=877.12±212.583,PFS=603.76±253.661) were longer than those of unfavorable group(OS=334.77±126.792,PFS=235.84±119.002)(P<.001).Radiomic and combined models with RF algorithm performed better than those with LR algorithm(AUCRF: 0.79,0.98;AUCLR: 0.51,0.51).The accuracy,sensitivity,specificity,and an area under ROC(AUC)for predicting favorable outcome with combined RF model(0.950,0.957,0.941,0.995) were higher than those with radiomic RF model(0.883,0.880,0.888,0.974) and clinical RF model(0.833,0.864,0.794,0.935).Conclusion: The combined model,which integrated radiomic and clincal factors,holds a potential better marker than radiomic and clinical models in predicting survival states in patients with HGG after resection and chemoradiotherapy.Part 2 MR features for predicting 1p/19 q codeletion status in lower grade glioma patientsObjective: The loss of chromosomal 1p/19 q heterozygosity(LOH)in lower grade glioma(LGG)is significantly correlated with chemotherapy response and prognosis.This study aims at predicting 1p/19 q LOH status in LGG patients by analyzing sample MR imaging features.Methods:The pre and postoperative MR and clinical data of 69 patients with LGG were retrospectively analyzed.The preoperative MR imaging features included tumor location,homogeneity,T2-FLAIR mismatch,subventricular zone(SVZ) involvement and midline shift.Postoperative MR features included morphology of new enhancement,range of FLAIR hyperintense around residual cavity,change of FLAIR signal intensity in the residual cavity and pseudo-progression.Clinical data included sex,age and WHO classification of tumor grade.The correlation between parameters and 1p/19 q LOH status was evaluated by univariate and multivariate logistic regression analysis.The sensitivity,specificity,PPV,NPV and accuracy of multivariate logistic regression model for predicting 1p/19 q LOH were calculated.Results:Univariate analysis showed that preoperative signs: tumor location(P<0.001),homogeneity(P=0.049),T2-FLAIR mismatch(P<0.001) and SVZ involvement(P=0.031) and postoperative signs: morphology of new enhancement(P<0.001),pseudo-progression(P=0.026) were correlated with 1p/19 q LOH.There were no statistical significant differences in midline shift,change of FLAIR signal intensity in the residual cavity,range of FLAIR hyperintense around residual cavity and clinical factors.Multivariate logistic regression analysis showed that homogeneity,T2-FLAIR mismatch and morphology of new enhancement were related to 1p/19 q LOH.The sensitivity,specificity and accuracy of multivariate logistic regression model for predicting 1p/19 q LOH were 0.74,0.83 and 0.80,respectively.Conclusion:This study assessed the correlation between conventional sample MR featuers and 1p/19 q gene status,and provided a noninvasive method for predicting 1p/19 q LOH.