A delta radiomics model based on ultrasound images predicts response to neoadjuvant therapy in triple negative breast...

Breast cancer is a common malignancy in women worldwide, with triple negative breast cancer (TNBC) being a particularly aggressive subtype. Current methods for assessing neoadjuvant therapy (NAT) response are often delayed, limiting timely adjustments to therapy. Delta radiomics offers a promising non-invasive approach to predict treatment outcomes by analyzing imaging changes over time.

A retrospective analysis was conducted on 101 female patients with TNBC who underwent NAT. A total of 972 delta radiomic features were extracted from ultrasound images acquired both pre- and post-NAT. T-test and least absolute shrinkage and selection operator (LASSO) were applied to select features for delta radiomics model development. A combined model was constructed by integrating the delta radiomics model with independent predictors. Receiver operating characteristic (ROC) curves, calibration curve, and decision curve analysis (DCA) were used to assess the predictive efficacy, calibration, and net clinical benefit of the models, respectively.

Multivariate regression analysis revealed that change rate of size (delta size) (odds ratio [OR] 2.74; p = 0.003) and Adler grade (pre-NAT) (OR 0.21; p = 0.030) were independent factors that influenced the prediction of pathologic complete response (pCR). Nine delta radiomics features were identified as significant and a delta radiomics model was subsequently developed. The combined model, which incorporates the delta radiomics model, delta size, and Adler grade, demonstrated an area under the curve (AUC) value of 0.850 (95% confidence interval [CI] 0.752-0.947) in the training cohort and 0.787 (95% CI 0.588-0.986) in the validation cohort. The calibration curves demonstrated that the combined model exhibited high calibration. The DCA showed a substantial net benefit across a range of clinically relevant risk thresholds.

The delta radiomics model based on ultrasound images has good predictive value for predicting pCR after NAT in TNBC and has the potential for clinical application.