The Research On The Prediction Of Microsatellite Instability In Rectal Cancer Based On MRI Radiomics Nomogram

Purpose:To investigate the feasibility of nonomogram combined with multi-parameter MRI radiomics and clinical factors for preoperative prediction of microsatellite instability in rectal cancer.Materials and Methods:The clinical and imaging data of 121 rectal cancer patients who underwent preoperative rectal MR examination and postoperative pathologically examination from March 2016 to March 2021 at the Affiliated Hospital of Qingdao University were retrospectively analyzed.All these patients did not receive preoperative radiotherapy,but all underwent immunohistochemical testing or polymerase chain reaction testing after the operation.All cases were randomly divided into a training set(n=85)and a validation set(n=36);the former makes up 70%of total cases,while the latter 30%.(1)MR images of transverse T1WI,T2WI and DWI were imported into 3D Slicer software by two experienced abdominal imaging diagnosticians.The tumor regions of each sequence were manually delineated layer by layer to generate 3D regions of interest and radiomics features were extracted.The single factor feature screening as well as least absolute shrinkage and selection operator algorithm(LASSO)were used for feature screening and dimensionality reduction,and the LASSO-Logistic method was adopted to construct seven radiomics models,consisting of multiple single-sequence,double-sequence,and triple-sequence radiomics models,to plot the receiver operating characteristic curve(ROC curve)of the test subjects,and to calculate the area under the curve(AUC)for evaluating the predictive performance of the seven radiomics models.(2)The comparison between microsatellite stability in rectal cancer and microsatellite instability in it in terms of clinical factors showed statistically significant differences(P<0.05).Clinical factors associated with the microsatellite instability were screened using univariate and multifactorial logistic regression analysis,and several clinical models were constructed.The clinical factors here included age,gender,the stage of rectal cancer,location of tumour,circumferential cutting edge,the mean ADC value of tumor area,The length of the tumor,the tumor’s degree of differentiation,smoking history,history of alcohol consumption,family history of tumor,CEA,CA125,CA199,red blood cell count in the blood,white blood cell count,platelet count,cholesterol level,high-density lipoprotein,and low-density lipoprotein.(3)The radiomics nomogram was created by using both the screened clinical factors and radiomics features.The receiver operating characteristic curve(ROC)was plotted and the area under the curve(AUC)was calculated to assess the predictive performance of the three models:clinical model,three-sequence radiomics model,and radiomics nomogram.Delong test is used to compare whether there is a statistical difference in the prediction efficiency of the three models.The calibration curve was plotted to assess the goodness of fit of the radiomics nomogram,and the Hosmer-Lemeshow testing was performed.Decision curve analysis(DCA)was used to assess the clinical application value of the nomogram.Results:(1)A total of 1130 radiomics features were retained in each sequence for each patient,and after single factor feature screening and dimensionality reduction by the LASSO algorithm,the respective radiomics features were retained in the training set,including 20 items in the DWI model,23 items in the T1 model,33 items in the T2 model,8 items in the DWI+T1 model,6 items in the DWI+T2 model,6 items in the T1+T2 model,and 8 items in the DWI+T1+T2 model.In the comparison between the seven radiomics models,the DWI+T1+T2 model showed the best predictive efficacy both in the training set[AUC value of 0.915(95%CI,0.851-0.967)]and in the validation set[AUC value of 0.872(95%CI,0.746-0.964)].In the comparison between two-sequence models,the T1+T2 model showed the best predictive efficacy.The predictive efficacy of the DWI+T1 model was comparable to that of the DWI+T2 model.In the comparison between single-sequence models,the DWI model showed the lowest predictive efficacy.(2)The results of univariate and multifactorial logistic analysis showed that platelet and high-density lipoprotein are clinical independent risk factors for predicting microsatellite instability(P<0.05),which were used to construct the clinical model in this study.The AUC values of the clinical model in the training and validation sets were 0.784(95%CI,0.682-0.885)and 0.725(95%CI,0.536-0.880),respectively.(3)The radiomics nomogram,consisting of platelet,high-density lipoprotein and radiomics score,outperformed the clinical model and the three-sequence radiomics model in terms of the predictive efficacy,in both the training set[AUC value of 0.966(95%CI,0.932-0.991)]and the validation set[AUC value of 0.931(95%CI,0.841-0.994)].The results of Delong test showed that there was a statistical difference between the predictive efficacy of the radiomics nomogram/the three-sequence radiomics model and that of the clinical model(P<0.05).The calibration curves of radiomics nomograms show good goodness of fit.The results of DCA show that the net benefit of nomogram is the largest.Conclusions:(1)The radiomics monogram created by combining multi-parameter MRI radiomics and clinical factors,as a noninvasive tool,is conductive to preoperatively predict the microsatellite instability of rectal cancer.(2)The radiomics monogram outperformed the clinical model and the three-sequence radiomics model in terms of the predictive efficacy.