Differential diagnosis of postoperative recurrence of triple-negative breast cancer based on two-dimensional ultrasound ra-diomics features of margin area of the lesion
Objective To investigate the value of preoperative marginal area radiomics analysis in the differential diagnosis of postoperative recurrence of triple-negative breast cancer(TNBC).Methods This study retrospectively collected data from 139 patients who underwent TNBC surgical resection at Haian People's Hospital from January 2016 to June 2022.Based on follow-up results,the patients were divided into recurrence and non-recurrence groups.All patients underwent breast two-dimensional ultrasound(2DUS)examination.By using Image J,the marginal area of the lesions was delineated on high-frequency grayscale images to extract the region of interest(ROI).Radiomics features were extracted using Pyradiomics.Intraclass correlation coefficient(ICC),univariate analysis,least absolute shrinkage and selection operator(LASSO)algorithm,and autocorrelation matrix were used for feature dimensionality reduction.Pearson correlation analysis was conducted to examine the relationship among radiomics features,Ki-67 index,and recurrence time.A logistic regression model was constructed based on radiomics features,and the model's differential diagnostic performance was evaluated using receiver operating characteristic(ROC)curves,calibration curves,and decision curves.Results After excluding 11 patients lost to follow-up,a total of 128 patients were included.Divided into a traning set of 90 cases and a testing set of 38 cases in a 7:3 ratio.Comparison between groups showed that five key features(X1:original_glszm_GrayLevelNonUniformity,X2:original_glcm_Contrast,X3:wavelet-HLL_gldm_LowGrayLevelEmphasis,X4:wavelet-HHH_glszm_GrayLevelNon Unifor-mity,X5:wavelet-HHH_glrlm_Gray LevelNonUniformityNormalized)had statistically significant differences(P＜0.05)after dimensionality reduction of marginal area radiomics features in the training set.Correlation analysis confirmed that X1,X2,and X4 were positively correlated with the Ki-67 index,and negatively correlated with recurrence time(all P＜0.05),while X3 and X5 were negatively correlated with the Ki-67 index,and positively correlated with recurrence time(all P＜0.05).Logistic regression model indicated that X2(OR=1.126,95％CI:1.086-1.165),X3(OR=1.100,95％CI:1.056-1.143),and X5(OR=1.142,95％CI:1.109-1.172)are independent influencing factors for TNBC recurrence.The model achieved an AUC of 0.892(95％CI:0.828-0.962)in the training set and an AUC of 0.873(95％CI:0.809-0.943)in the validation set.The calibration and decision curves demonstrated that the model has good calibration degree and clinical applicability.Conclusion Radiomics feature analysis on marginal area of the lesion has good differential diagnostic performance,and can effectively predict the risk of postoperative recurrence of TNBC.
Triple-negative breast cancerMargin areaImaging omicsRecurrenceUltrasound
国家科技期刊平台
NSTL
万方数据
