Exploring the Diagnostic Value of Ultrasound Examination for Uterine Leiomyoma and Adenomyosis Based on Radiomics
Objective This study aims to explore the application value of ultrasound radiomics in the differential diagnosis of uterine leiomyoma and adenomyosis.Methods A retrospective analysis of 150 cases diagnosed as uterine leiomyoma or adenomyosis by postoperative pathological examination in Xinzheng Maternal and Child Health Hospital from January 2020 to December 2022,and transvaginal color Doppler ultrasound examination was performed before treatment.According to the ratio of 7:3,it was randomly divided into a training set of 105 cases and a validation set of 45 cases.High-throughput data collection,feature extraction and dimension reduction were performed on the region of interest(ROI)on the ultrasound image,including data on gray scale,texture,shape,edge,etc.Then,a predictive model based on multifactor logistic regression was constructed,and multiple diagnostic efficiency parameters were used to evaluate the predictive effect of the model.Finally,the predictive model was validated.Results Among the 1,317 extracted feature parameters,11 radiomics feature variables related to the differentiation of uterine leiomyoma and adenomyosis were screened out.In the training set,the area under the curve(AUC)of the predictive model for distinguishing uterine leiomyoma and adenomyosis was 0.950(95%CI:0.909～0.990),with an accuracy of 91.41%,sensitivity of 77.44%,specificity of 97.21%,positive predictive value of 92.30%,negative predictive value of 71.22%,and cutoff value of 0.669.The AUC of the validation set was 0.854(95%CI:0.719～0.919),with an accuracy of 80.00%,sensitivity of 92.33%,specificity of 75.00%,positive predictive value of 60.00%,negative predictive value of 96.00%,and cutoff value of 0.474.Conclusion The results of this study confirm the feasibility and effectiveness of ultrasound radiomics in distinguishing uterine leiomyoma and adenomyosis,providing a simple,low-cost,and non-invasive screening method for clinical practice.
国家科技期刊平台
NSTL
万方数据
