Predictive Study Of HIFU Curative Effect Based On MRI T2WI Radiomics Features Of Uterine Fibroids

BackgroundThe efficacy of high-intensity focused ultrasound(HIFU)in the treatment of uterine fibroids is affected by many factors.The efficacy prediction has always been one of the main challenges faced by the clinical application of HIFU technology.The main reasons are: 1)The therapeutic effect is affected by the histopathology of uterine fibroids Differences such as the number of smooth muscle cells in fibroids and the content of collagen fibers,but still lack feasible evaluation indicators before surgery;2)Qualitative classification based on T2 signal intensity cannot provide accurate quantitative information for evaluating tumor heterogeneity for prediction The curative effect of HIFU has limitations;3)The difference in acoustic channel organization and fibroids position,type,and size of different patients have an important influence on the curative effect of HIFU,but there is no corresponding evaluation basis.Radiomics is the use of computer-assisted software to convert medical images into high-dimensional quantitative information,extract microstructure differences that cannot be recognized by the human eye,quantify tumor heterogeneity,and provide a quantitative and objective observation method for the body’s pathological changes.This article first carried out the correlation study of MRI T2WI radiomics features of uterine fibroids with the number of smooth muscle cells and collagen fiber content of fibroids,on this basis,considering the differences in acoustic channel organization,fibroids location and size of different patients,combining the patient’s clinical characteristics and radiomics features,and constructing HIFU ablation rate and Energy Efficiency Factor(EEF)prediction models based on machine learning to carry out the prediction research of HIFU curative effect.The results are expected to provide an objective and effective method for clinically quantifying tumor heterogeneity,predicting the efficacy of HIFU,and boosting the promotion of HIFU technology.Objective1.Carry out the correlation study of MRI T2WI radiomics features of uterine fibroids with the number of smooth muscle cells and collagen fiber content of uterine fibroids,and obtain the radiomics features that characterize the histopathological characteristics of uterine fibroids.2.Based on the MRI T2WI radiomics features of uterine fibroids,the ablation rate prediction model was established through machine learning to explore the feasibility of using radiomics features to predict HIFU ablation rate;radiomics features combined with clinical features to establish a combined radiomics-clinical model Further improve the accuracy of the forecast.3.Combining the MRI T2WI radiomics and clinical features of uterine fibroids,the energy efficiency factor prediction model is established through machine learning.Method1.The object of the study on the correlation between radiomics features and the content of smooth muscle cells and collagen fibers of uterine fibroids was performed in the First Affiliated Hospital of Chongqing Medical University from March 2011 to August 2016.Twenty-five cases of uterine fibroids whose fibroids were analyzed histopathologically.Collect preoperative MRI T2WI images of the patient,delineate the ROI layer by layer along the edge of the fibroids;extract the radiomics features of the ROI area;use Pearson or Spearman correlation coefficient analysis to evaluate the correlation between the radiomics features and the number of smooth muscle cells and the content of collagen fibers.2.223 patients were included in the study to predict the ablation rate of HIFU treatment of uterine fibroids.The 122 patients with uterine fibroids treated by HIFU in the First Affiliated Hospital of Chongqing Medical University were selected as the training set,46 patients as the internal validation set,and 55 patients from Chongqing Haifu Hospital as the external validation set.Based on the immediate postoperative ablation rate as a predictive indicator,patients were divided into two groups based on the 80% ablation rate(group H: ≥ 80%;group L: <80%).851 radiomics features were extracted from preoperative MRI T2WI images.The least absolute contraction and selection operator(LASSO)is used to select features in the training queue,and the support vector machine algorithm(SVM)is used to establish ablation rate prediction radiomics model,clinical model and radiomics-clinical combined model.Rad-score is calculated to explore the feasibility of radiomics features to predict ablation rate,area under the curve(AUC)is calculated to evaluate the predictive performance of different models,and decision curve analysis(DCA)is performed to evaluate clinical applicability.3.Using energy efficiency factor(EEF)as a predictor,the EEF prediction study was conducted on 223 patients with uterine fibroids treated with HIFU from the First Affiliated Hospital of Chongqing Medical University(n=168)and Chongqing Haifu Hospital(n=55).851 radiomics features were extracted from preoperative MRI T2WI images.Through correlation analysis,the radiomics features of high correlation with EEF were obtained,combined with clinical features,the EEF prediction model was established through linear regression.The model is evaluated by the correlation analysis between the predicted EEF value and the actual EEF value.Results1.851 radiomics features are extracted from MRI T2WI images.Through correlation analysis,a total of 5 features are highly correlated with the number of smooth muscle cells and collagen fiber content.Among them,GH＿10Percentile had the highest correlation with the number of smooth muscle cells(r = 0.547,P = 0.005);GLCM＿Autocorrelation had the highest correlation with collagen fiber content(r =-0.713,P = 0.000).2.Through feature selection,the 7 best prediction subsets are used to build the photoomics model.Rad-score shows that radiomics features have a certain value in predicting the ablation rate.The accuracy of the training set: 77.1%(94/122),the accuracy of the internal validation set: 73.9%(34/46),and the accuracy of the external validation set: 69.1 %(38/55).The AUC values of the internal validation set and the external validation set of the radiomics model established based on the radiomics features are0.793 and 0.767,respectively;the AUC values of the clinical model in the internal and external validation set are 0.705 and 0.707,respectively;the radiomics features The AUC values of the radiomics-clinical combined model established in combination with clinical features in the internal and external validation sets were 0.820 and 0.791,respectively.The decision curve analysis also shows that within the most reasonable threshold probability range,the radiomics-clinical combined model has the highest overall net benefit in predicting the ablation rate.3.8 radiomics features and 4 clinical features that are highly correlated with EEF were used to establish the EEF prediction model.In the internal validation set(n=46)and external validation set(n=55),the correlation between the predicted EEF value of the radiomics model and the actual EEF value was 0.417 and 0.579,respectively;the correlation between the predicted EEF value of the clinical model and the actual EEF value They are 0.683 and 0.687,respectively;the radiomics-clinical combined model established by combining radiomics and clinical features increases the correlation to 0.785 and 0.743,respectively.Conclusion1.The MRI T2WI radiomics features of uterine fibroids have a high correlation with the number of smooth muscle cells and the content of collagen fibers.It can provide a potential means for characterizing the histopathological characteristics of uterine fibroids,and improving T2 signal intensity is used to assess the limitations of uterine fibroids heterogeneity.2.Compared with the pure radiomics model and clinical model,the radiomics-clinical combined model based on radiomics and clinical features has a higher AUC value for predicting HIFU ablation rate,and it can be used as an objective and effective method to predict the ablation rate,helping clinicians to select patients with uterine fibroids who are most likely to benefit from HIFU treatment.3.Compared with radiomics model and clinical model,the EEF prediction model based on MRI T2WI radiomics and clinical features has higher correlation,and it can provide a new method for EEF prediction to evaluate the difficulty of HIFU treatment of patients with uterine fibroids.