The Value Of Radiolics And Amide-proton Transfer Imaging In Identification Of Biain High-grade Glioma And Metastatic Tumor

ObjectiveTo evaluate the efficacy of radioimcs and APT in differentiation high-grade glioma from metastatic tumor.Materials and methodsThis study retrospectively analysised 182 patients(91 glioblastomas and 91 cases of metastatic tumors)with conventional MRI and 31 patients(intracranial metastases and high grade glioma)with APT in January 2017-December 2019 in Shandong Medical imaging Research Institute.Radiomics characteristics of the brain tumor body and edema area were evaluated in T1WI,T2WI,T2FLAIR,CE-T1WI sequence.All patients were divided into training group(glioblastoma and metastatic tumor,71 cases)and validation group(20 cases).Random forest algorithm to feature selection and build the model of training was used.The validation group data were used validate the model.Kolmogorov-Smirnov test was used to verify the normal distribution of data.The difference of the normal distribution data was tested by T test Student,and the non-normal distribution variables were tested by Mann-Whitney U test.Spearman test was used to evaluate the association between different features.Area assessment under the work characteristic curve and the area under the subject operating characteristic curve was used to evaluate the predictive efficacy of the model for glioblastoma and metastatic tumor.For 31 patients who were examined with APT scan and routine images delineated the regions of interest(ROI)of tumor core area,peritumoral edema area and contralateral normal white matter area.The APTR values and MTR values of these three regions were measured and constructed pseudo-color maps.The distinguishing ability of APTR values and MTR values for high-grade gliomas and metastatic tumors was compared withworking characteristic curves of subjects.ResultsAmong the conventional sequence(T1WI,T2WI,T2-FLAIR,CE-T1WI).ten features with the highest importance were extracted from the radiomics features in tumor core area and peritumoral edema area(CORE-TRA_original_shape_MajorAxisLength?CORE-FLAIR_original_firstorder_TotalEnergy?CORE-FLAIR_original_shape_Maximum3DDiameter?CORE-FLAIR_exponential_glrlm?CORE-FLAIR_original_shape_?CORE-FLAIR_original_shape_Maximum2DDiameterRow?CORE-T2_logarithm_gldm_DependenceNonUniformity?CORE-T2_log-sigma-4-0-mm-3D_glszm_GrayLevelNonUniformity?CORE-T2_wavelet-LL_gldm_GrayLevelNonUniformity?CORE-T1_original_shape_Maximum2DDiameterColumn).All 1 0 features were derived from tumor core regions.The most important single feature Major Axis Length(original_shape_MajorAxisLength)derived from CE-T1WI,and five features derived from the T2-FLAIR sequence(40.5%).The importance of shape feature and texture feature was dominant among feature types(47.3%,45.7%).After the model was constructed,the AUC of the subject's operating characteristic curve in the training group was 0.85±0.14,and that in the verification group was 0.84.The MTR values of peritumor edema in high-grade glioma and metastatic tumor were 23.62±3.90 and 19.86±5.27,respectively.There isstatistically significant differences.The AUC value was 0.723(p=0.035),the 95%confidence interval 0.540-0.905,the threshold value 18.55,the sensitivity 92.9%,the specificity 52.9%,and the yoden index 0.458.ConclusionThe radiomics with MR conventional sequence could distinguished glioblastoma and metastatic tumor.APT is helpful in distinguish high-grade glioma from metastatic tumor.