Development and validation of a delta ultrasomics model for predicting treatment response to neoadjuvant chemotherapy...

Breast cancer is highly heterogeneous, it is imperative to determine patients who may benefit from neoadjuvant chemotherapy (NAC). This study aimed to develop and validate a delta radiomics-based model to predict the efficacy of neoadjuvant therapy in patients with breast cancer.

A retrospective cohort of breast cancer patients undergoing NAC was analyzed, with ultrasomics features extracted from ultrasound images taken before treatment and after two cycles. The delta ultrasomics features were scored as 0 or 1 based on whether expression changed between posttreatment and baseline. Treatment response was assessed based on the residual cancer burden score. The Chi-squared test combined with least absolute shrinkage and selection operator (LASSO) regression was implemented for feature reduction. Six machine learning algorithms were then used for model development. The area under curve (AUC), calibration curve and decision curve analysis (DCA) were used to evaluate model performance.

A total of 669 breast cancer patients were included in this study, and 468 and 201 patients were randomly separated into training and test sets at a ratio of 7:3. For the 1,239 extracted radiomics features, the Chi-squared test followed by the LASSO model ultimately included 25 delta ultrasomics features for model construction. For the six machine learning models, the gradient boosting machine (GBM) had the highest average AUC value in the training and test sets, which had remarkable predictive ability in the training set [AUC =0.912, 95% confidence interval (CI): 0.887-0.936] and test set (AUC =0.872, 95% CI: 0.824-0.920). Calibration and DCA curves revealed that the eight prediction models all performed well in predicting the response to NAC.

This study shows that delta ultrasomics model could be a valuable tool for predicting the response to NAC in breast cancer.