The Value Of Radiomics Based On MRI For T Stage And Predicting The Expression Of P53 In Rectal Cancer

Part Ⅰ The value of radiomics based on MRI for T stage inrectal cancerObjective:To construct multiparametric radiomics model based on high-resolution T2-weighted imaging（HRT₂WI）and diffusion-weighted imaging（DWI）with a b-value of800 to assess the the T-stage of rectal cancer,and non-invasively assess the progression of the patient’s disease.Materials and Methods:Retrospective analysis of 123 patients with rectal cancer who underwent 3.0T MRI at our hospital from January 2019 to December 2021.The clinical data collected included age,gender,tumor location,and tumor differentiation degree.According to the pathological T-stage results,patients with T1 and T2 stages were categorized as the non-breaking myeloid group,and patients with T3 and T4 stages were categorized as the breaking myeloid group.Fifty-seven cases in the unbroken myeloid group and 66 cases in the broken myeloid group were obtained.The features were extracted by two abdominal diagnosticians after manually outlining the VOI of the lesions,followed by feature dimensionality reduction using independent sample t-test and SVM linear function.The selected samples were randomly divided into training and validation sets according to 7:3,and support vector machine（SVM）classifier models were constructed to obtain the area under the subject operating characteristic（ROC）curve（AUC）,sensitivity,specificity and accuracy of the training and validation sets to assess the diagnostic efficacy of different models in predicting staging.The differences in AUC of different models were compared by De Long test.Results:1142 texture features were extracted from HRT₂WI and DWI（b=800）images of the tumor tissue of each patient in each case.The features were screened by independent sample t-test,respectively.After screening 41 and 83 features from HRT₂WI and DWI respectively by independent sample t-test,the texture features with higher feature weights were selected for significance analysis,and the results showed that the texture features in HRT₂WI and DWI had significant differences in different T-stages.The SVM model constructed based on the texture features of HRT₂WI images had an AUC value of 0.904,sensitivity of 87.0%,specificity of 82.5%,and accuracy of84.9%in the training group.The validation group had an AUC value of 0.894,a sensitivity of 90.0%,a specificity of 70.6%,and an accuracy of 81.1%.The AUC value of the SVM model constructed from DWI image texture features for the training group was 0.880,with a sensitivity of 82.6%,specificity of 82.5%,and accuracy of 82.6%.The validation group had an AUC value of 0.774,sensitivity of 60.0%,specificity of76.5%,and accuracy was 67.6%.The final combined HRT₂WI and DWI prediction model was significantly more effective than the other models,with an AUC value of0.943,sensitivity of 91.3%,specificity of 85.0%,and accuracy of 88.4%in the training group.The validation group had an AUC value of 0.927,sensitivity of 80.0%,specificity of 88.2%,and accuracy of 83.8%.the De Long test showed a significant difference between the predictive efficacy of the SVM model with combined HRT₂WI and DWI and the sequence model alone（P<0.05）.Conclusions:1.SVM model based on texture features of HRT₂WI and DWI can effectively evaluate T-stage of rectal cancer.2.The radiomics model combining HRT₂WI and DWI enable higher predictive efficacy than individual sequence models,facilitating personalized patient treatment.Part Ⅱ The value of radiomics based on MRI for predicting the expression of p53 in rectal cancerObjective:To construct machine learning models including support vector machine（SVM）,random forest（RF）and logistic regression（LR）based on HRT2 WI images to evaluate their value in predicting p53 gene expression in patients with rectal cancer.Materials and Methods:The image data of 90 rectal cancer patients who underwent HRT2 WI in our hospital from January 2019 to December 2021 were retrospectively analyzed.The clinical data of patients were collected including age,gender,tumor location,degree of tumor differentiation and T stage.Patients were categorized into 32 cases in the negative group and 58 cases in the positive group according to the immunohistochemical p53 expression status of postoperative pathology.Textural features were extracted by two abdominal diagnosticians after manually outlining the VOI of the lesions,followed by feature downscaling using independent sample t-test.The selected samples were randomly divided into training and validation sets in the ratio of 7:3,and SVM,RF,and LR classifier models were constructed to obtain the area under the working characteristic（ROC）curve（AUC）,sensitivity,specificity,and accuracy of subjects in the training and validation sets to assess the diagnostic efficacy of different models in predicting staging.The differences in AUC of different models were compared by Delong test.Results:1142 texture features were extracted from the HRT2 WI images of tumor tissues of each patient,and the features were screened by independent sample t-test and linear classifier.Thirty-three features were screened by t-test and SVM linear function,and the texture features with higher feature weights were selected for significance analysis,which resulted in different p53 expression states wavelet LHL_firstorder_Mean,logsigma50mm3D_glcm_Imc1,wavelet LLLL_glcm_Contrast and other texture parameters have significant differences.The SVM model constructed from the texture features of HRT2 WI images had an AUC value of 0.936,sensitivity of 87.5%,specificity of 77.3%,and accuracy of 83.9% for the training group.The validation group had an AUC value of 0.717,a sensitivity of 77.8%,a specificity of 60.0%,and an accuracy of 71.4%.The constructed RF model training group had an AUC value of 0.996,sensitivity of 95.0%,specificity of 99.7%,and accuracy of 96.8%.The validation group had an AUC value of 0.639,a sensitivity of 77.8%,a specificity of 50.0%,and an accuracy of 67.9%.The AUC value of the constructed LR model training group was 0.955,with a sensitivity of 95.0%,a specificity of 77.3% and an accuracy of 88.7%.The validation group had an AUC value of 0.728,sensitivity of 83.3%,specificity of 60.0%,and accuracy of 75.0%.the De Long test demonstrated that the differences in AUC values between the different models were not statistically significant（P>0.05）.there was a significant difference in the degree of tumor differentiation between p53 negative and positive expression（P<0.05）,and the degree of differentiation of negative was higher than that of positive expression.Conclusion:The radiomics based on HRT2 WI can effectively predict the p53 expression of rectal cancer.There is no significant differences in predictive efficacy among the three imagingomics models（SVM,RF,LR）.All of them have the potential to non-invasively analyze important biomarkers of rectal cancer,which may help with clinical treatment and prognosis assessment.