Clinical Study On Predicting Complete Remission (CR) Of Radiotherapy And Chemotherapy In Patients With Cerivcal Cancer By Radiomics

Objective To compare the texture information of pre-treatment MRI images of cervical cancer patients with different therapeutic effects,analyze the different features,establish a prediction model and verify its accuracy.Method A total of 80 patients with cervical cancer who underwent concurrent radiotherapy and chemotherapy in our hospital from May 2018 to October 2020 were collected.All patients underwent conventional magnetic resonance and dynamic contrast-enhanced magnetic resonance examination before and after treatment,and were divided into CR group and non-CR group according to whether there were tumor residues after the end of radiotherapy and chemotherapy cycle.The MRI images of different groups of patients before treatment were exported from PACS system in bmp format.The MRI image analysis software Mazda was used to delineate the tumor area layer by layer to obtain the three-dimensional information of tumor,and the tumor analysis function was run to obtain more than 100 groups of image characteristic parameter information.Pathon,an advanced computer language,is used to import data and standardize and randomize the data.T-test(test standard p<0.05)is used to compare two independent samples,and features with differences are screened out.Lasso feature screening is used to calculate the coefficient of variables with little correlation with the results as 0.Three prediction models are established based on the screened features: logical regression model,random forest model and support vector machine model,and 50%crossover is used.The receiver operating characteristic curve(roc)of the prediction model was drawn,and its prediction efficiency AUC was calculated.The confidence interval was 95%,and the difference was statistically significant with p<0.05.The imaging score R_score of patients can be obtained by calculating all screening features multiplied by their coefficients,summing up and adding their constants.the R_score of each patient is calculated and combined with the general clinical data of patients.according to the treatment effect,the significance sig is calculated,showing that sig<0.05 has significant difference,while sig>0.05 has no significant difference.Result 1.According to the t-test and lasso feature screening,seven groups of imaging omics features with predictive performance are obtained:Z_GLev Non U,Z_Fraction,Perc.99%3D,Gr Skewness,S(1,0,0)Correlat,S(0,1,0),Contrast Entropy.2.Compared with the prediction model established by the selected features,logistic regression,random forest and support vector machine,it is found that the support vector machine has higher accuracy and stability by 50%cross-validation.3.The significance of R_score established by lasso is sig=0.02<0.05,and the sig values of other clinical data are all greater than 0.05.Conclusion1.the characteristic information of cervical cancer patients was analyzed by magnetic resonance imaging,which can predict the PCR after radiotherapy and chemotherapy;2.In this experiment,the support vector machine model has better reliability and stability than other models.3.Compared with other clinical information,the R_score of omics established by the characteristic parameters of imaging omics is the only index that can predict the curative effect.