Font Size: a A A

Identification Of WHO Grade 4 Glioma In Midline Region And Prediction Of H3 K27 Alteration Based On Multimodal MRI

Posted on:2024-07-20Degree:MasterType:Thesis
Country:ChinaCandidate:S H LiangFull Text:PDF
GTID:2544307133997899Subject:Imaging and nuclear medicine
Abstract/Summary:PDF Full Text Request
Background and purposeGlioma is the most common primary brain tumor,accounting for approximately 80%of primary malignant tumors in the central nervous system(CNS).Gliomas occurring in the midline structures of the brain(corpus callosum,thalamus,midbrain,and pontine)and the cerebral hemispheres(frontal,temporal,parietal,occipital,and insular lobes)have different histological features and genetic molecular phenotypes,among which H3 K27 is a key molecular marker for the classification and diagnosis of gliomas in the midline region.It was found that substitution of lysine at position 27 in the H3 histone gene by methionine leads to reduced methylation of the histone tail and inhibition of polycomb repressive complex 2(PRC2)methyltransferase function,which in turn prevents glial cell differentiation and promotes glioma formation.Depending on the status of H3 K27 and IDH,WHO grade 4 gliomas in the midline region can be classified as 1)H3 K27 mutation with IDH wild: diffuse midline glioma,H3 K27 altered(Diffuse Midline Glioma,H3 K27 Altered,a-DMG);2)H3 K27 wild and IDH wild: IDH wild type Glioblastoma(IDH-wildtype,w-GBM).Gliomas in the midline region occur with H3 K27 mutations in up to 80% of children and adolescents,and the proportion of mutations in this gene is higher than that in adults.a-DMG and w-GBM are currently usually intervened with a standardized treatment strategy of maximum safe range surgical resection and concurrent radiotherapy,but there are significantly different survival and prognosis.Relevant targeted and immunosuppressive therapies based on changes in H3 K27 status are essential to improve survival expectations of patients.It is worth noting that histopathological test results are not available for all gliomas in the midline region in clinical work.On the one hand,tumors located in the deep midline region of the brain make invasive surgical resection and puncture biopsy techniques risky and difficult to be tolerated by some patients;on the other hand,even if surgical resection is performed,postoperative pathological tissue obtained only represents a limited area of the tumor,and the high degree of heterogeneity of the glioma itself makes pathological analysis possible with false-negative results.On the other hand,even if surgical resection is performed,the pathological tissue obtained after surgery only represents a limited area of the tumor,and the high heterogeneity of glioma itself makes the pathological analysis may have false-negative results.In contrast,magnetic resonance imaging(MRI),as a non-invasive diagnostic method,can comprehensively analyze the signal characteristics of tumors and their surrounding tissue structures,and is widely used for qualitative and quantitative evaluation of brain tumors.The current conventional and functional MRI have an important role in identifying different histological types of brain tumors and predicting the molecular phenotype of specific genes.Studies related to conventional MRI have shown that circumferential enhancement,peritumoral edema,cysts,and T2WI-FLAIR mismatches can help predict altered H3 K27 gene status.However,the problems are that the diagnostic efficacy of individual morphological features is low;the interobserver agreement of some features is relatively poor;the imaging features of gliomas in the midline region are not consistent with those of gliomas in the cerebral hemispheres,and some VASARI features are not suitable for describing gliomas in the midline region.The results of functional MRI-related studies showed that functional MRI parameters such as apparent diffusion coefficient based on diffusion-weighted imaging and cerebral blood flow based on perfusion-weighted imaging were statistically different between the two groups.However,the problems are that the diagnostic cutoff values or diagnostic models are not yet uniform between studies;the accuracy and reproducibility of quantitative parameters need to be improved.Therefore,reliable imaging techniques are necessary to obtain accurate quantitative parameters.With the advancement of imaging technology,Multi-delay Arterial Spin Labeling(MD ASL),which is based on traditional arterial spin labeling,has been applied in the diagnosis and treatment of stroke and epilepsy.The advantage of MD ASL is that,compared to conventional arterial spin labeling,multiple post-labeling delays(PLDs)can be set during the acquisition process,which can improve the accuracy of conventional ASL parameters;more important parameters can be obtained to reflect blood perfusion,such as blood flow and arterial passage time;different PLD settings can better quantify the perfusion characteristics of different tissue types of tumors;In addition,arterial spin labeling imaging does not require the introduction of exogenous contrast agents and is more suitable for patients with renal insufficiency or contrast allergy.Therefore,the identification of WHO grade 4 gliomas in the midline region and the prediction of H3 K27 gene status based on conventional MRI morphological features and MD ASL perfusion parameters need further study.Materials and Methods:In the first part,WHO grade 4 glioma identification and H3 K27 status prediction in the midline region of the brain based on conventional MRI features.Clinical baseline data,conventional MRI morphological features were collected from patients with surgically pathologically confirmed a-DMG and w-GBM in the midline region of the brain.A total of 108 cases with a-DMG(45 cases)and w-GBM(63 cases)were finally included,and the108 patients were divided into a training set(76 cases,a-DMG/w-GBM = 31/45)and a test set(32 cases,a-DMG/w-GBM = 14/18)according to a 7/3 ratio.All MRI sequence images were anonymized and four neuroimaging attending physicians collected two types of tumor conventional MRI features based on VASARI and custom MRI features,and the histopathological findings of the tumors were not known during the analysis of the images.All MRI features were analyzed for intergroup variability in the training set,test set,and total set,respectively.One-way logistic regression analysis was performed for features that were clinically significant or statistically different with good interobserver agreement(Kappa/ICC values >0.75).VASARI features that contributed significantly(P < 0.05)to the predicted outcome were screened using the forward likelihood ratio method of multifactor logistic regression.Subsequently,a multifactor regression model was constructed based on the custom image features and VASARI features.Continuous variables were statistically analyzed using independent sample t-tests or nonparametric tests(Mann-Whitney U test).Categorical variables were statistically analyzed using the chi-square test or Fisher exact test.The diagnostic efficacy of each model was evaluated by calculating the area under the curve(AUC)from the subject operating characteristic curve(ROC).The De Long test was used to compare whether there was a difference in AUC between the models.Calibration curves were plotted to evaluate the model fitting efficacy in the training and test sets.Apply decision curve analysis to evaluate the net benefit and clinical usefulness of the models.Visualize the best predictive model using a nomogram and visualize and quantify the individual features in the model.In the second part,a prediction model was constructed for grade 4 gliomas in the midline region of the brain based on MD ASL perfusion parameters with conventional MRI features.Seventy-eight patients with WHO grade 4 gliomas in the midline region of the brain confirmed by surgical pathology and undergoing MD ASL were collected,including a-DMG(32 patients)and w-GBM(46 patients).CBF and TDT-related parameters were measured by two attending neuroimaging physicians in 78 tumors with MD ASL,and the histopathological findings of the tumors were not known to anyone during the analysis of the images,and the mean of the two physicians’ measurements was taken for statistical analysis.Using post-processing software(GE AW4.6 workstation)to outline the region of interest(ROI)in the CBF map,to measure the blood flow values in the tumor high perfusion region(CBFmax),the blood flow values in the low perfusion region(CBFmin),the difference between the blood flow values in the high and low perfusion regions(CBFmm),and the blood flow values in the normal contralateral frontal brain region(CBFnorm),and to calculate the relative values of each perfusion parameter;The ROI was outlined in the TDT map using the same method to obtain TDT-related parameters(TDTmean and TDTnorm),and the relative passage delay time of the tumor(r TDTmean)was calculated;the volume share of the hyperperfused area within the tumor(PHPA)was observed.Finally,a multifactorial logistic regression model was constructed based on MD ASL perfusion parameters and conventional MRI morphological features.Continuous variables were statistically analyzed using independent samples t-test or non-parametric test(Mann-Whitney U test);categorical variables were statistically analyzed using chi-square test or Fisher exact test.The diagnostic efficacy of each model was evaluated by calculating the area under the curve(AUC)from the subject operating characteristic curve(ROC).The De Long test was used to compare whether there was a difference in AUC between the models.Calibration curves were plotted to evaluate the model fitting efficacy.Apply decision curve analysis to evaluate the net benefit and clinical usefulness of the models.Visualize the best predictive model using a nomogram,and visualize and quantify the individual features in the model.Results:Results in Part I: Of the 108 patients,45 had a-DMG(30.7±19.5 years)and 63 had w-GBM(49.8±16.9 years),and age was statistically different between the two groups(P< 0.001).A total of 21 features were statistically different in the total set(P < 0.05).The AUC values obtained for the custom image feature model in the training and test sets were0.883(95% CI: 0.789,0.945)and 0.925(95% CI: 0.774,0.988),respectively;the AUC values obtained for the VASARI feature model in the training and test sets were 0.918(95% CI: 0.0.833,.0.969)and 0.893(95% CI: 0.805,1.000);the AUC values obtained for the joint custom image features and VASARI feature model were 0.921(95% CI: 0.849,0.977)and 0.930(95% CI: 0.904,1.000)in the training and test sets,respectively.In the joint model,age,MDT and MERS(yes)were independent predictors of H3 K27 mutation(P < 0.05).the De Long test showed that in the test and training set,the AUC values of the3 models were not statistically different.Decision curve analysis showed that all 3 models had good clinical benefit.A nomogram was constructed based on the joint model with the highest diagnostic efficacy,with parameters including T2 FM,PN,IMS,MERS,SEM,MDT,and age.Results in the second part: age,CBFmax,CBFmm,r CBFmax,r CBFmm,TDTmean,r TDTmean,and percentage of hyperperfusion were statistically different between the two groups(P < 0.05).The diagnostic efficacy of CBFmax was the highest among the perfusion parameters with an AUC value of 0.749(95% CI: 0.640,0.858)and a sensitivity and specificity of 66.7% and 64.3%,respectively.The perfusion parameter model had an AUC value of 0.892(95% CI: 0.807,0.967),with sensitivity,specificity and diagnostic accuracy of 83.3%,80.9% and 84.2%,respectively;the conventional MRI feature model had an AUC value of 0.925(95% CI: 0.864,0.986),with sensitivity,specificity and diagnostic accuracy of 86.1%,83.3% and 86.8%,respectively The Delong test showed that the AUC values of the combined perfusion parameter and conventional MRI feature models were 0.945(95% CI: 0.899,0.991),and the sensitivity,specificity and diagnostic accuracy were 83.3%,90.5%,and 89.5%,respectively.In the perfusion parameter model,PHPA(<50%),r CBFmax,and r TDTmean were risk factors for the development of a-DMG,and r CBFmm was a protective factor for the development of a-DMG.The joint model showed that MERS(yes),r CBFmax,and r TDTmean were independent predictors of H3 K27 mutation status(P < 0.05).A nomogram was constructed based on the joint model with the highest diagnostic efficacy,with parameters including r CBFmax,r CBFmin,r TDTmean,PHPA,MERS,and Age.Conclusion:Part I: 1)A multiparametric model constructed based on conventional MRI morphological features can be used to identify a-DMG and w-GBM and predict the H3K27 gene status of grade 4 gliomas in the midline region;2)There is no significant difference in diagnostic efficacy between custom MRI feature models and VASARI feature models,and custom models are more concise;3)Combining custom features and VASARI features can improve the diagnostic efficacy of the combined model.Part II: 1)A multiparametric model based on perfusion parameters of MD ASL can be used to identify a-DMG and w-GBM and predict H3 K27 gene status in grade 4gliomas in the midline region;2)the combined perfusion parameters and custom conventional MRI feature model have the highest diagnostic efficacy;3)MER sign is an independent predictor in both parts of the study.
Keywords/Search Tags:Brain midline structure, WHO 4 grade Glioma, H3 K27 altered, Magnetic resonance imaging, VASARI, MD ASL
PDF Full Text Request
Related items