| HCC is one of the most common primary liver malignant tumors,and its incidence is increasing all over the world,which is the second leading cause of cancer-related death in China.Surgery and liver transplantation are the main treatment strategies for patients with HCC,but the poor prognosis caused by high rate of recurrence and metastasis is the main cause of death.Histological differentiation and the expression of Ki-67 are important factors reflecting the invasiveness and prognosis of HCC.The histological differentiation and Ki-67 expression of HCC need to be obtained by surgical resection or preoperative puncture biopsy in clinic.For patients,when identified as a certain pathological type of HCC,they often miss the best time for treatment and are prone to some unavoidable complications.Therefore,scholars at home and abroad have put forward a large number of studies on the non-invasive prediction of biological characteristics of HCC by radiomics models.However,most studies focus on the radiomics features of intratumor,but ignore the clinical value of tumor microenvironment in predicting the biological characteristics of HCC.In recent years,some scholars have begun to pay attention to the influence of peritumoral radiomics features on the performance of model,and confirmed that peritumoral radiomics features can further improve the differentiation of benign from malignant lesions and the prediction of biological characteristics in breast,colon,lung and liver.While,in the extraction of peritumoral radiomics features,the selection of region is still controversial.For the liver cancer,some researchers choose 1cm or 2cm outside the tumor as the peritumoral area,but do not take the size of the tumor into account.As result,this study attempts to explore the application value of different ranges of peritumoral radiomics features in the prediction of biological characteristics of HCC.Determining the best peritumoral region of HCC by comparing the performance of models,so as to extract the best peritumoral radiomics features,improve the performance of the radiomics model,and further enrich the content of the radiomics model.This study is mainly divided into two parts.In the first part,the prediction model of HCC differentiation degree was constructed based on the radiomics features of intratumor and peritumor.We extracted the radiomics features of different peritumoral areas to construct multiple models,and selected the optimally peritumoral region through comparing the performance of the models.At the same time,we discussed the influence of the number of layers of the region of interest on the performance of the model.Based on the optimally peritumoral range selected in the first part,radiomics models were constructed to predict the level of Ki-67 expression in HCC in the second part.We discussed the feasibility of peritumoral radiomics features and the optimal peritumoral region by comparing the performance of intratumoral radiomics model,peritumoral radiomics model and combined radiomics model.[Methods]1.Clinical informationThe preoperative CT images of 200 patients with pathologically proved HCC was collected in our hospital from January 2016 to January 2019.The pathological results of all the patients contained clear information of the degree of HCC differentiation,of which 104 patients were examined by immunohistochemistry,including a clear positive index of Ki-67.Preoperative CT images included non-enhanced phase,arterial phase,portal venous phase and delayed phase.The interval between preoperative CT examination and operation or puncture was no more than one month.Grouping:In the first part,according to the degree of differentiation of HCC,patients with HCC were divided into well to moderate differentiation group and poor differentiation group.According to the number of CT layers,it can be divided into 2D ROI(maximum diameter layer of tumor)and 3D ROI(maximum diameter layer of tumor and the upper and lower two layers).Based on the positive index value of Ki-67 in HCC,patients were divided into Ki-67 high expression group(Ki-67 positive index more than 10%)and Ki-67 low expression group(Ki-67 positive index less than 10%)in the second part.2.Evaluation methodThe AUC value,accuracy,sensitivity and specificity of each model were calculated respectively.3.Statistical analysisSPSS20.0 and Medcalc19.4.1 software were used for statistical analysis.All the data were tested by normal distribution test and variance homogeneity test.The continuous variable of normal distribution were represented by X±S,the continuous variable of non-normal distribution were represented by average value and range,and the classified variable were expressed by percentage.The continuous variable of the two groups were analyzed by two independent sample t-test,and the classified variables were compared by chi-square test.Areas under the curve for different models were compared using the DeLong test.A P-value of<0.05 was considered to indicate statistically significant differences.[Results]The first part:1.Construction and validation of the prediction model of HCC differentiation based on intratumoral and peritumoral dynamic-CT radiomics featuresIn the validation dataset,the AUC value,accuracy,sensitivity and specificity of 3D intratumoral radiomics model were 0.64(95%CI:0.65-0.72),0.70,0.58 and 0.75 respectively,while those of 2D intratumoral radiomics model were 0.68(95%CI:0.56-0.79),0.73,0.48,0.88 respectively.The AUC value of 3D optimally peritumoral radiomics model was the best in the peritumoral 1.3-1.4 times region,which was 0.76(95%CI:0.64-0.86),and the AUC value of 2D optimally peritumoral radiomics model was 0.80(95%CI:0.69-0.89)in the peritumoral 1.2-1.3 times region.The optimal AUC values of 2D and 3D radiomics models were 0.85(95%CI:0.74-0.92)and 0.80(95%CI:0.68-0.88)in the peritumoral 1.3-1.4 times region.The performance of the optimally combined model was better than that of the optimally peritumoral radiomics model,and was significantly better than the intratumoral radiomics model(0.85 vs 0.68,P=0.009,0.80 vs 0.64,P= 0,74)and the clinical model(0.85 vs 0.62,P=0.02,0.80 vs 0.62,P=0.16).2.The influence of 2D ROI or 3D ROI on the performance of the model for predicting the differentiation of HCCIn the models of intratumoral radiomics,optimal peritumoral radiomics and optimally combined radiomics,the performance of 2D radiomics features in evaluating the differentiation of HCC was better than that of 3D radiomics features((0.68 vs 0.65,P=0.71),(0.80 vs 0.76,P=0.54),(0.85 vs 0.80,P=0.32)),but there were no statistical difference.The AUC values of 3D and 2D optimally peritumoral radiomics model were the best in the peritumoral 1.2-1.3 times region and 1.3-1.4 times region,respectively.The AUC values of 2D and 3D optimally combined model were the best in the peritumoral 1.3-1.4 times region.The second part:1.Construction and validation the prediction model of Ki-67 expression level based on intratumoral and optimally peritumoral regionIn the validation dataset,the AUC value,accuracy,sensitivity and specificity of 3D intratumoral radiomics model were 0.70(95%CI:0.51-0.85),0.59,0.60 and 0.83 respectively,while those of 2D intratumoral radiomics model were 0.80(95%CI:0.62-0.92),0.81,0.85 and 0.75,respectively.The AUC values of 3D and 2D peritumoral radiomics models were 0.67(95%CI:0.48-0.83)and 0.87(95%CI:0.70-0.96),respecttively.The AUC values of 2D and 3D combined radiomics models were 0.90(95%CI:0.75-0.98)and 0.78(95%CI:0.59-0.90),respectively.The performance of the combined radiomics model was better than that of the peritumoral radiomics model,and was significantly better than that of the intratumoral radiomics model(0.90 vs 0.80,P=0.07,0.78 vs 0.70,P=0.49)and the clinical model(0.90 vs 0.60,P=0.02,0.78 vs 0.60,P=0.27).2.The influence of 2D ROI or 3D ROI on the performance of model for predicting Ki-67 expression level in HCC.In intratumoral radiomics,peritumoral radiomics and combined radiomics models,the performance of 2D radiomics features in evaluating the expression level of Ki-67 in HCC was better than that of 3D radiomics features[Conclusion]1.The peritumoral radiomics model had the best performance in peritumoral 1.2-1.3 times or 1.3-1.4 times region,while the combined radiomics model achieved the best performance in peritumoral 1.3-1.4 times region.The optimally combined radiomics model was significantly better than the intratumoral radiomics model and the clinical model in predicting the differentiation of HCC.The add of peritumoral radiomics features significantly improves the predictive performance of the intratumoral radiomics model.2.The width of the 1.3-1.4 times region around the tumor selected in this study is about 0.7mm to 18mm.Different from other studies,the optimal peritumoral region we selected varies with the size of the lesion rather than a fixed value.3.Based on the optimal peritumoral region shown by the results of the first part,this part constructed and validated the models of intratumoral radiomics,peritumoral radiomics and combined radiomics model.The results showed that the combined radiomics model with peritumoral radiomics features improved the performance of the intratumoral radiomics model.The performance of combined radiomics model was better than that of intratumoral radiomics model and clinical model.It suggests that the peritumoral radiomics features are helpful to predict the expression level of Ki-67 in HCC.4.In the study of HCC differentiation and Ki-67 expression,we compared the performance of 2D and 3D models.The results of the two parts showed that the performance of 2D radiomics models were generally better than that of 3D radiomics model,although there were no statistical difference.It still suggests that we can consider using the maximum diameter of the tumor to extract radiomics features for a large amount of data in clinic. |
