A new prediction model of triple-negative breast cancer based on ultrasound radiomics

Background: This study aims to assess the diagnostic potential of a radiomics model based on ultrasonic imaging and characteristics for triple-negative breast cancer (TNBC). Methods: We retrospectively assessed the data of 127 patients with breast tumors, dividing them into training (n = 88) and testing (n = 39) sets. Four machine learning (ML) algorithms were employed, and we compared three distinct prediction models. Accuracy, sensitivity, specificity and the area under the receiver operating characteristic curve (AUC) served as the metrics for evaluating the predictive capacity of these models for TNBC. Results: Multivariate logistic regression analysis identified margin (odds ratio (OR) 0.296; 95% confidence interval (CI) 0.127–0.692; p = 0.005) and posterior echo (OR 0.323; 95% CI 0.112–0.930; p = 0.036) as independent TNBC predictors. We selected thirteen key image features to construct an ultrasonic imaging radiomics model. In the ultrasonic imaging radiomics model, the AUC values for logistic regression (LR), support vector machine (SVM), decision tree (DT), and random forest (RF) were 0.78, 0.80, 0.84 and 0.95 for the training set, and 0.68, 0.55, 0.71 and 0.65 for the testing set, respectively. In the ultrasonic feature model, the AUC values for LR, SVM, DT and RF were 0.73, 0.54, 0.73 and 0.73 for the training set, and 0.58, 0.82, 0.59 and 0.60 for the testing set, respectively. In the comprehensive model, which combines ultrasonic feature and radiomics feature models, the AUC values for LR, SVM, DT and RF were 0.85, 0.92, 0.78 and 0.92 for the training set, and 0.64, 0.69, 0.63 and 0.76 for the testing set, respectively. Conclusions: Compared with ultrasonic feature and radiomics feature models, the comprehensive model demonstrated better diagnostic performance and could identify TNBC more effectively.