Study On Preoperative Magnetic Resonance Imaging In Predicting 1p/19q Molecular Characteristics Of Glioma

Background:Glioma is the most common brain tumor in the adult central nervous system.The biological behavior of WHO grade II-III gliomas is different from that of WHO grade IV gliomas.It is classified as lower-grade gliomas（LGG）by the Cancer Genome Atlas（TCGA）.The molecular characteristics of chromosome arm 1p and 19q（1p/19q）play an important role in the classification diagnosis,treatment and prognosis evaluation of LGG,but the qualitative diagnosis of this molecular characteristics mainly depends on postoperative molecular pathology.How to use imaging methods to predict the 1p/19q molecular characteristics of glioma noninvasively before surgery is a hot issue in clinical research and practice.T2-Fluid attenuated inversion recovery（FLAIR）mismatch sign（T2-FLAIR mismatch sign）refers to the presence of T2WI hyperintensity areas（non-cystic or necrosis）in gliomas,most of which can be effectively suppressed by FLAIR,and a high signal ring can be seen.This sign has been proved to have good specificity in predicting LGG molecular typing,but it has low sensitivity and false positive.Moreover,the current image quantitative identification criteria for T2-FLAIR mismatch sign are not clear,which limits the clinical application value of this sign in predicting LGG molecular typing.Diffusion weighted imaging（DWI）can non-invasively evaluate the diffusion degree of water molecules in tumor tissues.Apparent diffusion coefficient（ADC）has been proved to be related to the pathological information of glioma,and has good diagnostic value in the grading,benign and malignant differentiation and molecular typing of glioma.However,previous studies on ADC value prediction of 1p/19q mainly establish prediction models by multi-point averaging.Due to the heterogeneity of gliomas,the accuracy and reliability of the prediction results need to be further improved.Radiomics can reflect the heterogeneity of glioma more comprehensively by extracting massive medical image information and exploring the internal characteristics of tumor,which can provide more accurate basis for evaluating molecular characteristics.As far as we know,there are few reports on the predictive value of ADC-based radiomics models for the 1p/19q molecular characteristics of glioma,especially the combination of T2-FLAIR mismatch sign and ADC radiomics models in the prediction of 1p/19q has not been reported.This study intends to focus on the scientific problem of accurate preoperative diagnosis of glioma 1p/19q,and use retrospective case studies to carry out the predictive value of preoperative magnetic resonance imaging in noninvasive prediction the 1p/19q molecular characteristics of glioma,including the following two parts:1.To verify the diagnostic value of T2-FLAIR mismatch sign in predicting the 1p/19q molecular characteristics of glioma,to further extract the quantitative parameters of T2-FLAIR mismatch sign,to explore whether the quantitative parameters can improve the predictive performance of T2-FLAIR mismatch sign,and to clarify the image quantitative identification criteria of T2-FLAIR mismatch sign.2.Based on the preoperative magnetic resonance ADC map,the radiomics score was established to predict 1p/19q molecular characteristics.Combined with T2-FLAIR mismatch sign,its diagnostic performance in predicting 1p/19q molecular characteristics was discussed.Materials and Methods:This is a single-center retrospective study.The clinical and imaging data of LGG patients with WHO grade II-III confirmed by postoperative pathology between January 2017 and December 2021 were collected.All patients underwent Siemens Verio 3.0T magnetic resonance scanner and 8-channel head coil before operation.The scanning sequences included T1WI,T2WI,FLAIR,enhanced T1-weighted imaging,DWI and susceptibility weighted imaging.It is divided into the following two parts:The main purpose of the first part is to verify the diagnostic performance of T2-FLAIR mismatch sign in predicting the 1p/19q molecular characteristics of glioma.According to the results of molecular pathological examination,the patients were divided into1p/19q-codeleted（1p/19q-Codel）group and 1p/19q-noncodeleted（1p/19q-Noncodel）group.T2-FLAIR mismatch sign and MR image features,including midline shift,subventricular zone（SVZ）involvement,cortex infiltrated,tumor margin,calcification,cystic or necrosis,peritumoral edema and tumor enhancement,were evaluated by two neuroradiologists in a blind manner.If two neuroradiologists independently judged that the above MRI features were different,they reached an agreement after consultation.In order to clarify the image quantitative identification criteria of T2-FLAIR mismatch sign,the patients were divided into mismatch sign positive and negative groups according to the results of MR image analysis.The quantitative analysis of T2-FLAIR mismatch sign was performed by hot spot method.Five circular regions of interest were delineated in the tumor area and the normal appearing white matter in contralateral semioval center,and the T2-FLAIR signal inhibition rate and relative mean apparent diffusion coefficient(r ADC_mean)were calculated.The main purpose of the second part is to establish an imaging prediction model of1p/19q molecular characteristics for T2-FLAIR mismatch sign combined with ADC radiomics and verify the diagnostic performance of the model.The LGG patients collected retrospectively were divided into training and verification sets by complete randomization according to the ratio of 7:3,and then the patients were divided into 1p/19q-Codel group and1p/19q-Noncodel group according to the results of molecular pathology.After image preprocessing,image segmentation was performed independently by a neuroradiologist.Thirty images were selected for segmentation between neuroradiologists to evaluate the stability of the extracted features.Volume of interest（VOI）was defined as an abnormal area in FLAIR image except obvious cystic or necrosis.The VOI extracted from the FLAIR image was copied to the r ADC image for radiomics feature extraction,and the features with good stability were retained for Z-score standardization.Pearson or Spearman correlation analysis and LASSO analysis were used for feature selection.The selected radiomics features were used to construct the radiomics label and calculate the score to establish the radiomics score（Rad-score）.The receiver operating characteristic（ROC）curve was used to evaluate the performance of the radiomics model and verified in the validation group.A predictive model of T2-FLAIR mismatch sign combined with Rad-score was established by multivariate logistic regression analysis and a nomogram was drawn.The area under curve（AUC）,calibration plot and Hosmer-Lemeshow test were used to evaluate the effectiveness of the prediction model,and internal validation was performed through the validation set.Statistical analysis:Chi-squared was used to analyze the differences in clinical and MRI features between 1p/19q-Codel and 1p/19q-Noncodel groups,and between training and validation sets.Kappa test was used to analyze the consistency of the T2-FLAIR mismatch sign and associated conventional MRI tumor image features between the two neuroradiologists.Univariable and multivariable logistic regression were used to screen independent predictors,and a prediction model was established and a nomogram was drawn.ROC curve,calibration plot and Hosmer-Lemeshow test were used to evaluate the performance of the model.Bootstrap method was used to internally verify the model.Mann-Whitney U test and independent sample t test were used to analyze the difference of T2-FLAIR inhibition rate and r ADC_mean value between mismatch positive and negative groups.ROC curve was used to evaluate the diagnostic performance of quantitative parameters of T2-FLAIR mismatch sign in differentiating positive and negative mismatch groups.Wilcoxon test was used to evaluate the difference of Rad-score between 1p/19q-Codel and 1p/19q-Noncodel groups in the training and validation sets.Intra-rater and inter-rater correlation coefficients（ICCs）were used to evaluate the stability of the features extracted by the two neuroradiologists.De Long test was used to evaluate the difference in AUC between the prediction models established by different factors.A two-tailed p value of<0.05 was considered to be statistically significant.Results:1.T2-FLAIR mismatch sign in predicting 1p/19q molecular characteristics of lower-grade gliomas1.1 A total of 146 LGG patients were included in this study,including 68 patients in1p/19q-Codel group and 78 patients in 1p/19q-Noncodel group.The incidence of T2-FLAIR mismatch sign in 1p/19q-Noncodel group（38.5%）was higher than that in 1p/19q-Codel group（16.2%）,and the difference was statistically significant（p<0.05）.In addition,the incidence of cortex infiltrated and calcification in the 1p/19q-Codel（77.9%and 38.2%,respectively）was higher than that in the 1p/19q-Noncodel group（57.7%and 12.8%,respectively）,and the difference was statistically significant（all p<0.05）.1.2 Univariable and multivariable logistic regression analysis showed that T2-FLAIR mismatch sign（p=0.007）,cortex infiltrated（p=0.001）,and calcification（p=0.003）were independent predictors of 1p/19q status.The AUC value of T2-FLAIR mismatch sign in predicting 1p/19q-Noncodel was 0.61,with a specificity was 83.82%and a sensitivity was38.46%.1.3 ROC curve analysis showed that the prediction model established by T2-FLAIR mismatch sign combined with cortex infiltrated and calcification had good performance,with an AUC value of 0.740,and the AUC value of internal verification by Bootstrap method was0.740.At the same time,the calibration and fitting of the model are good.The results of De Long test showed that T2-FLAIR mismatch sign combined with cortex infiltrated and calcification was more effective than T2-FLAIR mismatch sign in predicting 1p/19q molecular characteristics,and the difference was statistically significant（p=0.002）.1.4 In order to clarify the image quantitative identification criteria of T2-FLAIR mismatch sign,146 LGG patients were divided into 41 patients in mismatch positive group and 105 patients in mismatch negative group.The T2-FLAIR signal inhibition rate and r ADC_mean value of the mismatch positive group were higher than those of the mismatch negative group,and the difference was statistically significant（all p<0.001）.ROC curve analysis showed that the threshold of T2-FLAIR signal inhibition rate in the diagnosis of positive mismatch group was 44.26%,and its AUC value,accuracy,sensitivity and specificity were 0.975,93.15%,95.12%and 92.38%,respectively.The threshold value of r ADC_mean for the diagnosis of positive mismatch was 1.95,and the AUC value,accuracy,sensitivity and specificity were 0.911,83.56%,82.93%and 83.81%,respectively.2.T2-FLAIR mismatch sign combined with ADC radiomics model in predicting1p/19q molecular characteristics of lower-grade gliomas2.1 A total of 146 LGG patients were randomly divided into training set（n=102）and validation set（n=44）according to the ratio of 7:3.There was no evidence of difference in clinical and MR imaging features between the training set and the validation set（p>0.05）.2.2 A total of 1037 radiomics features were extracted from the VOI of r ADC maps,and the remaining 920 features were removed after the poor stability features were removed with ICC>0.75 as the standard.Through Pearson or Spearman correlation analysis and LASSO regression analysis,15 non-zero coefficient features were selected and Rad-score was constructed.In the training and validation sets,there was a significant difference in the Rad-score between the 1p/19q-Codel group and the 1p/19q-Noncodel group（p<0.001）,and the Rad-score model showed good predictive performance（AUC values were 0.896 and 0.778,respectively）.2.3 ROC curve analysis showed that the AUC values of T2-FLAIR mismatch sign in the training and validation sets were 0.611 and 0.631,respectively.The prediction model of T2-FLAIR mismatch sign and Rad-score showed good predictive performance in the training and validation sets,with AUC values of 0.899 and 0.754,respectively.In addition,De Long test results showed that T2-FLAIR mismatch sign combined with Rad-score was more effective than single T2-FLAIR mismatch sign in predicting 1p/19q.Conclusion:1.T2-FLAIR mismatch sign can be used as a specific imaging sign to predict1p/19q-Noncodel LGG.T2-FLAIR mismatch sign combined with cortex infiltrated and calcification can improve the predictive efficacy of this sign.The quantitative parameters of T2-FLAIR mismatch sign are helpful to clarify the image recognition criteria of T2-FLAIR mismatch sign.2.The Rad-score based on ADC maps can noninvasively predict the 1p/19q status of LGG.T2-FLAIR mismatch sign combined with ADC Rad-score model can improve the performance of T2-FLAIR mismatch sign in predicting 1p/19q molecular characteristics.