Prediction Of Pathological Classification,ablation Difficulty And Immediate Ablation Rate Of Uterine Fibroids Treated By HIFU Based On Conventional MRI And T2WI–radiomics Features

Part 1 Diagnostic value of MR-T₂WI based radiomics in distinguishing different pathological types of uterine fibroidsBackground Uterine fibroid is the most common benign type of uterine tumors in women of childbearing age.It can cause a series of symptoms,such as frequent urination,urgent urination,prolonged menstruation and pelvic pain,which seriously affects the quality of life of women.Cellular type,degenerated type and ordinary type are the three most common pathological types of uterine fibroids.Different pathological types of uterine fibroids have different treatment methods and prognosis.It has a great clinical significance to identify different types of uterine fibroids before treatment correctly.MRI is the most important noninvasive evaluation method before treatment of uterine fibroid.Different pathological types of uterine fibroid have some signal differences on MRI according to the difference of T₂WI and DWI signal,the types of uterine fibroid can be distinguished to a certain extent based on T₂WI and DWI,but the T₂WI and DWI signals of different types of uterine fibroid have partial overlap.Radiomics reflects tissue heterogeneity and biological characteristics by mining high-throughput data behind the conventional images.It has been applied in many organs of the body and proved to have important clinical value.However,the application of radiomics parameters in the pathological classification of uterine fibroid is relatively less,which is worthy of further studies.Objective To summarize the signal characteristics of T₂WI and DWI in different pathological types of uterine fibroid,and to explore the diagnostic value of conventional imaging features and radiomics characteristics in distinguishing different pathological types of uterine fibroids,focusing on the identification of cellular uterine fibroids.Methods The clinical and conventional MR data of 63 cases（65 uterine fibroids）confirmed by surgical pathology between March 2015 and June 2020 were collected,including 20 cases（21 lesions）of ordinary uterine fibroids,22 cases（23 lesions）of cellular type,21 cases（21 lesions）of degenerated type.Lesions segmentation and features extraction were performed by ITK-SNAP and A.K.software on T₂WI respectively.The Pearson correlation,f-SBF（random forest function）and 10 fold cross validation sampling were used to select the features from the extracted radiomics features.Then,the train function in caret package of R language was used to train the selected training group and construct a conditional inference tree model.ROC curve and confusion matrix were used to calculate the overall diagnostic performance of the model.Finally,65 cases of uterine uterine fibroids were divided into two groups: cellular uterine fibroids group and other group.The conventional MR features with statistical difference and the selected radiomics features with important diagnostic value were chosen for subsequent multivariate logistic regression analysis.The diagnostic efficiency of conventional MRI model,radiomics model and combined model were compared.Results 1.There were statistical differences in the component ratio of T₂WI signal and DWI signal among the three subtypes of uterine fibroids.The homogeneous high signal ratio was the most（65.2%）in the cellular type uterine fibroids group,and the mixed high signal was the most in the degenerated type（61.9%）on T₂WI.The high signal ratio of cellular type uterine fibroids was higher than the degenerated type and the ordinary type（P < 0.001）.There was no significant difference in age,location and type of uterinefibroids among the three groups（F = 1.17,P = 0.317）.There were also significant differences in the signal component ratio of T₂WI signal and DWI signal between cellular type and other two types of uterine fibroids（P < 0.001）.2.A total of 828 features were extracted from each lesion,and 12 of them were selected as the most valuable features for differential diagnosis among the three types of uterine fibroids.The AUC of the conditional inference tree model established based on these features to identify ordinary type,cellular type and degenerated type of uterine fibroids were 0.97,0.82 and 0.91,the sensitivity were 100%,70.6% and 80%,the specificity were 93.8%,90%,90.6%,the positive predictive value were 88.2%,80%,80%,and the negative predictive value were 100%,84.4%,90.6% respectively.The AUC of validation group were 0.92,0.73 and 0.78,the sensitivity were 85.3%,50%,66.7%,the specificity were 100%,75%,75%,positive predictive value were 100%,50 %,57.1 %,and negative predictive value were 92.3 %,75 %,82 %,respectively.3.The area under the curve（AUC）of conventional MRI model were 0.909（95% CI: 0.812;0.966）,the accuracy was 86.96%,the sensitivity was 82.61%,the specificity was 95.24%,the positive predictive value was 90.5%,and the negative predictive value was 90.9%;the AUC of radiomics model was 0.954（95% CI: 0.872;0.991）,the accuracy was 91.3%,the sensitivity was 91.3%,and the specificity was 97.62%,the positive predictive value was 95.5%,and the negative predictive value was 95.3%;the AUC of the combination model was 0.997（95% CI: 0.939;1）,the accuracy,sensitivity,specificity,positive predictive value and negative predictive value of the combination model were 95.65%,95.65%,100%,100% and 97.7%,respectively.Conclusion 1.According to T₂WI and DWI,different pathological types of uterine leiomyoma can be identified to some extent,but there is overlap between MR signals among them.2.The three classification model based on conditional inference tree can effectively distinguish the cellular type,degenerated type and ordinary type of uterine fibroids.Among which the diagnosis efficiency of ordinary uterine fibroids is the highest and the cellular type is the lowest.3.Single radiomics features and conventional MR features can not establish an effective differential diagnosis model.The combination of radiomics and conventional MR features can establish a differential diagnosis model with high model fitting and accuracy,and the diagnostic value is higher than single conventional MRI or radiomics.Part 2 The Predictive value of conventional MR features combined with radiomics features in energy efficiency factor（EEF）of HIFU ablation uterine fibroidsBackground It is very necessary to evaluate uterine fibroids with difficulty ablation by HIFU and amount of ultrasound energy needed for treatment.EEF is one of the most accurate quantitative indicators of HIFU ablation,which reflects the energy deposition efficiency of HIFU.EEF prediction model based on conventional MRI has a certain clinical application value,but conventional MRI is based on qualitative imaging,there are some overlaps between different signal intensities,and the human eye is often unable to recognize them.Radiomics can extract massive features from conventional imaging images to reflect the subtle differences within tissues.It has potential value in the prediction of EEF,which is worthy of further study.Objective The aim of this study was to establish EEF prediction models of HIFU ablation uterine fibroids with conventional MRI features,conventional MR and T₂WI-radiomics features,and to explore the influencing factors of EEF in the prediction model.Then compare the two models,so as to explore the additional value ofradiomics features on conventional MRI further.Methods A total of 216 symptomatic uterine fibroids in 216 female patients were treated with HIFU therapy from October 2015 to March 2020.The baseline clinical and MR parameters before and after HIFU ablation were retrospectively analyzed.And the EEF was calculated according to the above results.Lesions segmented and features extracted were performed by ITK-SNAP and A.K.software on T₂WI respectively.The minimum redundancy and maximum（mRMR）were used to select the radiomics features,and 20 features with high correlation but no redundancies with NPVR were retained.SPSS software was used to establish multiple linear regression models using the conventional MR features,conventional MRI and radiomics combined features,respectively.The NPVR related features of the two models were found out,and the prediction efficiency of the two models were compared,statistically.The validation of the predictive effectiveness of the final model was based on the correlation analysis between the predicted EEF value and the actual EEF value.Results 1.The results of conventional MRI parameter model showed that T₁ WI enhancement degree(X₁₁: mild = 0,moderate = 1,severe = 2)and T₂WI signal intensity（X₉: hyporintense signal = 0,iso-intensesignal = 1,hyperintense signal = 2）of uterine fibroids;the location of hysteromyoma（X₄: anterior wall = 0,posterior wall = 1,lateral wall = 2）had positive affect on EEF,and the size of uterine fibroids（X₂: expressed by the diameter）had negative affect on EEF.The regression equation was EEF = 12.110 + 4.261X₁₁﹣2.067X₂+3.868X₉+3.371X₄。2.Conventional MRI and radiomics features combined model showed that wavelet＿HHH＿firstorder＿Skewness(X₁₂),T₁ WI enhancement(X₁₁: mild = 0,moderate = 1,marked = 2),T₂WI signal intensity（X₉: hyporintense signal = 0,iso-intensesignal = 1,hyperintense signal = 2）and DWI signal intensity（X₁0: hyporintense signal = 0,iso-intensesignal = 1,hyperintense signal = 2）had positive effect on EEF,original＿shape＿Maximum2DDiameter Slice(X₁₃)had negative affect on EEF.The regression equation is: EEF = 14.901 + 58.177 X₁₂ + 3.753 X₁₁﹣0.26X₁₃+ 3.089 X₉ + 2.828 X₁0.3.The adjusted R² of conventional MRI model and combined model were 0.144 and 0.297 respectively,and the two fitted model was statistically significant（P < 0.05）.The Durbin Watson values of the two models were 2.007 and 1.958,respectively.The predicted EEF value of the combined model was 9.61 [3.60;15.22] J / mm³,and the actual EEF value was 6.24 [3.40;11.01] J / mm³,with the correlation coefficient r = 0.5.Conclusion 1.The conventional MR features,the combination of conventional MRI and radiomics features can establish a reasonable prediction model of NPVR for uterine fibroids ablation by HIFU.2.Conventional MRI model showed that EEF was positively correlated with T₁ WI enhancement,T₂WI signal intensity and the location of uterine fibroids,and negatively correlated with the maximum diameter（Dmax）of uterine fibroids..3.The combined model showed that wavelet＿HHH＿firstorder＿Skewness,T₁ WI enhancement,T₂WI and DWI signal intensity had positive affect on EEF,and_original＿shape＿Maximum2DDiameter Slice had negative affect on it.4.The prediction efficiency of the combined model is better than that of the conventional MRI model,and the radiomics parameters have important supplementary value for conventional MRI.There was moderate correlation between the predicted EEF of the combined prediction model and the actual EEF,which has a certain popularization value.Part 3 The Predictive value of conventional MR features combined with T₂WI-radiomics in immediate ablation rate of HIFU ablation uterine fibroidsBackground NPVR of HIFU ablation for uterine fibroids is an important parameter to evaluate the therapeutic effects,which is closely related to the long-term outcomes.MRI is the most important method used in preoperative evaluation and selecting cases for uterine fibroids treated with HIFU.At present,the preliminary curative effect prediction and ablation difficulty judgment of uterine fibroids were based on conventional MR features in clinical trials,and the NPVR prediction model has been established based on conventional MRI.Previous research showed that conventional MRI has important value in predicting NPVR of HIFU ablation of uterine fibroids.Radiomics features have a high predictive value than conventional MRI in identifying tissue heterogeneity and subtle differences.It has been widely used in the diagnosis,differential diagnosis,efficacy prediction and other aspects of the whole body.However,there is no report about radiomics features on the establishment of NPVR prediction model after HIFU ablation for uterine fibroids,so the value of radiomics features in NPVR treated by HIFU deserves further investigation.Objective The aim of this study was to establish NPVR prediction models of HIFU ablation uterine fibroids with conventional MRI features,conventional MR and T₂WI-radiomics features respect,and to explore the influencing factors of NPVR in the prediction model.Then compare the two models,so as to explore the additional value of radiomics features on conventional MRI.Methods A total of 216 symptomatic uterine fibroids in 216 female patients were treated with HIFU therapy from October 2015 to March 2020.The baseline clinical and MRI parameters before and after HIFU ablation were retrospectively analyzed.And theEEF was calculated according to the above results.Lesions segmented and features extracted were performed by ITK-SNAP and A.K.software on T₂WI respectively.The minimum redundancy and maximum（mRMR）were used to select the radiomics features,and 20 features with high correlation but no redundancies with NPVR were retained.SPSS software was used to establish multiple linear regression models using the conventional MR features,conventional MRI and radiomics combined features,respectively.The NPVR related features of the two models were found out,and the prediction efficiency of the two models were compared,statistically.The validation of the predictive effectiveness of the final model was based on the correlation analysis between the predicted NPVR value and the actual NPVR value.Results 1.The results of conventional MR features model showed that the of T₂WI signal intensity（X₉: hyporintense signal = 0,iso-intensesignal = 1,hyperintense signal = 2）,T₁ WI enhancement degree(X₁₁: mild = 0,moderate = 1,marked = 2),and the location of uterine fibroids（X₄: anterior wall = 0,posterior wall = 1,lateral wall = 2）had negative effects on NPVR.The multiple linear regression equation is NPVR =103.851﹣11.868X₉﹣5.64X₁₁﹣2.984X₄.2.Conventional MRI and radiomics features combined model showed that the T₂WI signal intensity（X₉: hyporintense signal = 0,iso-intensesignal = 1,hyperintense signal = 2）,T₁ WI enhancement degree(X₁₁: mild = 0,moderate = 1,marked = 2),the location of uterine fibroids（X₄: anterior wall = 0,posterior wall = 1,lateral wall = 2）,glszm＿Size Zone Non Uniformity(X₁₂)firstorder had negative effects on NPVR.The regression equation is: NPVR=104.03﹣11.886×X₉﹣5.459X₁₁﹣2.776X₄﹣0.20X₁₂﹣16.913X₁₃3.The adjusted R² of conventional MRI model and combined model were 0.385 and 0.408 respectively,and the fitted model was statistically significant（P < 0.05）.There was no col-linearity between the parameters.The Durbin Watson value of the model was close to the standard value 2.The predicted NPVR value of the combined modelwas 81[71;91] %,and the actual NPVR value was 89[77;97] %,with the correlation coefficient r = 0.655（P<0.001）.Conclusion 1.The conventional MR features,the combination of conventional MRI and radiomics features can establish a reasonable prediction model of NPVR for uterine fibroids ablation by HIFU.2.Conventional MRI model showed that the T₂WI intensity signal,T₁ WI enhancement degree and the location of uterine fibroids had negative effects on NPVR.The combined model showed that T₂WI signal intensity,T₁ WI enhancement degree,location of uterine fibroids and wavelet＿HHL＿glszm＿Size Zone Non Uniformity 、wavelet＿HHH＿firstorder＿Skewness had negative effects on NPVR.3.The combined model has a slightly greater predictive power than the conventional MRI model in predicting NPVR after HIFU ablation for uterine fibroids.Radiomics has a certain complementary value to conventional MRI,but the value is limited.There was moderate correlation between the predicted NPVR of the combined prediction model and the actual NPVR,which has clinical practical value in the treatment of uterine fibroids by HIFU.