MRI-based Radiomics For Preoperative Prediction Of Ki-67 Status For Tongue Squamous Cell Carcinoma

ObjectiveTo explore the value of preoperative prediction of ki-67 status for tongue squamous cell carcinoma(TSCC)based on MRI radiomics.Materials and methodsAll of the 56 patients with tongue squamous cell carcinoma(TSCC)confirmed by postoperative pathology were retrospectively analyzed in China-Japan Union Hospital of Jilin University from January 2016 to October 2020,and divided into "high expression group"(27 cases)and "low expression group"(29 cases)according to ki-67≥30% and ki-67 < 30%.All of the patients underwent routine MR sequence scanning and DWI sequence scanning of head and neck before operation,and the corresponding image data and relevant clinical pathology data were perfected.Two radiologists used Dr.Wise multimodal scientific research platform to import T2WI and ADC images,and then ROI was delineated layer by layer along the edges of TSCC tumor parenchyma in these images,and then the radiomics features were extracted from the ROI.The intraclass correlation coefficient(ICC)was used to analyze the repeatability of the radiomics features extracted by the two observers from the MRI images,and the radiomics features with ICC < 0.75 were excluded.Subsequently,the L1 Based-LASSO method was used to screen out the 10 optimal radiomics features for the modeling verification.All clinical pathological data were examined by Spearman correlation analysis,univariate and multivariate logistic regression analysis.And the closely related clinical factors were selected and included in the clinical prediction model.Receiver operating characteristic curve(ROC)was used to evaluate the predictive efficacy among the radiomic models in each group,and calibration curves and decision curves(DCA)were used to evaluate the calibration efficacy and clinical applicability based on the radiomics model.ROC of diifferent models are compared by DeLong’s test.ResultsAll of the 56 patients with TSCC were included 47 males and 9 females with the average age of 57.48±8.84 years old.(1)Spearman correlation analysis and multivariate logistic regression analysis showed that T stage was an independent risk factor for ki-67 expression in patients with TSCC(OR=11.404;95%CI:1.245-104.422;P < 0.05),which was positively correlated with ki-67 expression(r=0.630,P < 0.01).(2)The ROC curve results show that the prediction performance of the combined model in the verification group is slightly better than that of the ADC model,and both of them have relatively excellent prediction abilities.The AUC values are 0.943 and 0.891,the specificity(SPE)is 94.6% and 89.7%,the sensitivity(SEN)is 74.1% and 81.5%,and the accuracy(ACC)is 87.5% and 85.7%respectively.The clinical model had low specificity and good sensitivity in the verification group,with the AUC of 0.791,sensitivity(SEN)of 92.6%,specificity(SPE)of 65.5%,and accuracy(ACC)of 78.6%.The T2WI model had moderate predictive abilities in the verification group,with an area under the curve(AUC)of0.831,sensitivity(SEN)of 85.2%,specificity(SPE)of 75.9%,and accuracy(ACC)of80.4%.(3)De Long’s test has revealed that the prediction efficiency of the combined model is superior to that of the T2WI model and the clinical model,and the prediction efficiency of the ADC model is superior to that of the clinical model.However,the significant differences in prediction efficiency have not been found between the combined model and the ADC model,between the ADC model and the T2WI model,and between the T2WI model and the clinical model.(4)Through the calibration curve and decision curve demonstrated that the combined model,ADC model and T2WI model have good calibration performance,and each cluster model has good net benefits under almost all threshold probabilities.Particularly,within the threshold probability range of 60%-80%,each cluster model basically shows good net benefits differences.The nomograms based on the combined model provide clinicians with an intuitive and easy tool to predict the status of ki-67 expression in patients with TSCC.ConclusionMRI-based radiomics can predict the preoperative expression of ki-67 in patients with TSCC,especially the combined model and ADC radiomics model.Combined model and ADC model have excellent prediction performance in specificity and accuracy.In addition,the nomogram constructed based on the combined model is expected to provide a non-invasive prediction tool for clinicians to make reasonable diagnosis and treatment decisions.