Font Size: a A A

MRI Based Radiomics And Texture Analysis In The Prediction Of Pathological Subtypes And Perineural Invasion In Rectal Cancer

Posted on:2024-04-08Degree:MasterType:Thesis
Country:ChinaCandidate:X F ZhouFull Text:PDF
GTID:2544306938980859Subject:Medical imaging and nuclear medicine
Abstract/Summary:PDF Full Text Request
Part Ⅰ:Application of texture analysis based on T2WI in predicting patients with perineural invasionObjective:To explore texture analysis in predicting perineural invasion of rectal cancer before surgery using T2 weighted imaging(T2WI)images.Methods:T2WI pictures.clinial and pathological information of 105 patients diagnosed with rectal adenocarcinoma were collected.Patients were divided into perineural invasion and non-perineural invasion group according to the presence or absence of perineural invasion.All patients underwent routine magnetic resonance imaging(MRI)before surgery.Texture features are obtained by delineating regions of interest on all slices of T2 weighted images,using Mazda.The differences in clinical.pathological and MRI texture features between the two groups were compared using two independent sample t test and Mann-Whitney test(P<0.05).The r-value(Spearman’s rank correlation coefficient)was used to calculate the correlation between texture features and perineural invasion status in patients.Receiver operating characteristic curve(ROC)was drawn for the parameters with statistical difference between the two groups,and the area under the curve(AUC)was analyzed.Using the binary logistic regression method to calculate the joint predictive value of each feature and calculate the AUC.The DeLong test was used to compare the difference of AUC between the multi-parameter texture features and the prediction curves based on single texture features.Correlations between texture features and perineural invasion status in patients were calculated using r-values(Spearman’s rank correlation coefficient).Results:There were no significant statistical differences of CEA and gender between two groups((P=0.061,0.487,0.097).Pathological T and N stage of tumor between two groups showed statistical difference.with P<0.05.Seven texture features from T2WI images of rectal cancer showed significant differences between two groups,namely:S(2.0.0)Sumaverg.S(0.0.2)Correlat.S(3.0.0)Sumaverg.S(0.0.3)AngScMom.S(0.0.3)Correlat.S(4.0.0)SumAverg,S(5,0.0)SumAverg.Perineural invasion status of rectal cancer and S(2.0.0)SumAverg.S(3.0.0)SumAverg.S(0.0.3)Correla.S(4.0.0)SumAverg.S(5.0.0)SumAverg showed negative correlation and were statistically different(P<0.05).and the r values were-0.206,-0.199,-0.208,-0.220,-0.225,respectively.S(5.0.0)SumAverg and S(4.0.0)SumAverg showed the AUC in predicting perineural invasion:0.642 and 0.649,respectively.The ROC curve showed that the joint performance of multiple texture features in predicting is better than the single texture feature.The AUC was 0.779.The combined performance in predicting of sensitivity and specificity are 64.47%and 85.71%,respectively.There is a statistical difference of the AUC between the joint prediction curve and each single prediction curve(P<0.05).Conclusion:Texture features based on T2WI images can predict the presence of perineural invasion in rectal cancer.And the joint performance of multiple parameters in predicting is better than that of the single texture feature.It is expected to be used for precise treatment and prognosis evaluation of rectal cancer patients.Part Ⅱ:Application of T2WI based radiomics in differentiating pathological subtypes of rectal cancerObjective:To study the value of radiomics in identifying non-signet ring cell and non-mucinous adenocarcinoma from mucinous adenocarcinoma of rectal cancer based on T2WI of MRI.Methods:T2-weighted imaging pictures,clinial and pathological information of patients with mucinous adenocarcinoma and non-signet ring cell and non-mucinous adenocarcinoma were retrospectively collected.50 patients were with mucinous adenocarcinoma and 66 patients were diagnosed as non-signet ring cell and non-mucinous adenocarcinoma.According to the ratio of 7:3,all patients were divided into the training group and the validation group.35 patients had mucinous adenocarcinoma and 46 patients were with non-signet ring cell and non-mucinous adenocarcinoma in the training group.There were 15 patients with mucinous adenocarcinoma and 20 patients with non-signet ring cell and non-mucinous adenocarcinoma in the validation group.The region of interest of rectal cancer was delineated by all slices on the T2WI imags.and then the radiomic features were obtained by using Mazda.The Least absolute shrinkage and selection operator(LASSO)method was used to extract the radiomic features and to obtain the radiomics score.The clinicopathological characteristics were analyzed whether there were statistical differences using two independent sample t test and Mann-Whitney test(P<0.05).The receiver operating characteristic curve of the radiomics score in the training group,.and the decision curve analysis(DCA)was used to evaluate the clinical practicability of the prediction model.The accuracy and reliability of the radiomics model was verified.Results:The radiomics score finally included 9 radiomics features:MaxNorm3D.Variance3D.Kurtosis3D,Perc.90%3D.Perc.99%3D.GrMean.S(2.-2.0)DifVarnc.S(0.0.2)SumVarnc.S(5.5.0)SumVarnc.The constructed LASSO formula is:y=4.156863*0.1-MaxNorm3D*1.247867*0.001+Variance3D*5.929337*0.0001+Kurtosis3D*2.258639*0.001+Perc.90%3D*2.965135*0.001+Perc.99%3D*5.788952*0.0001+ GrMean*4.572299*0.0001-S(2.-2.0)DifVarnc*1.639506*0.001S(0.0.2)SumVarnc-4.240341-0.00001+S(5.5.0)SumVarnc*1.493981*0.001.There were no significant differences in age,gender,stage,lymph node,CEA,perineural invasion,and vascular invasion status between two groups of patients.The AUC of the radiomics score in the training group was 0.653,the Youden index J was 0.2913.the sensitivity was 89.13%,and the specificity was 40.00%.The AUC of the radiomics score in the validation group was 0.657,the Youden index J was 0.4000.the sensitivity was 60%.and the specificity was 80%.The DCA curve shows that the model is useful for clinical decision-making.Conclusion:MRI-based radiomics can be used to differentiate non-signet ring cell and non-mucinous adenocarcinoma from mucinous adenocarcinoma.which is expected to guide personalized treatment of rectal cancer.
Keywords/Search Tags:Rectal cancer, Texture analysis, Perineural invasion, MRI, Prediction, Radiomics, Pathological subtype
PDF Full Text Request
Related items