CT-based radiomics analysis for prediction of response to neoadjuvant chemotherapy in breast cancer: A systematic review.

Breast cancer is the most commonly diagnosed cancer in women, and a major cause of cancer-related mortality. Neoadjuvant chemotherapy (NAC) is a key treatment for locally advanced breast cancer, but the response varies among patients. Radiomics extracts quantitative imaging features from computed tomography (CT), and thus offers a non-invasive approach to predict NAC response. This systematic review evaluates the predictive performance of CT-based radiomics models and their clinical applicability. Following PRISMA guidelines, a comprehensive literature search was conducted of PubMed, EMBASE, and Cochrane CENTRAL. Studies assessing CT-based radiomics for NAC response prediction in breast cancer were included. Key outcomes included area under the receiver operating characteristic (ROC) curve (AUC), accuracy, sensitivity, and specificity. Thirteen studies, prospective and retrospective, with a range of 39-549 patients were included. Machine learning models including LASSO regression and convolutional neural networks, were used in the studies. AUC values ranged from 64 % to 100 %, with deep learning models achieving superior performance. Hybrid models integrating clinical and radiomic features outperformed radiomics alone. Sensitivity and specificity varied, with deep learning models generally having better specificity, and hybrid models better sensitivity. In conclusion, CT-based radiomics shows promise in predicting NAC response in breast cancer, particularly with deep learning and hybrid models. However, variability in feature selection and validation methods highlight the need for standardization. Future research should focus on integrating multimodal imaging and clinical data to enhance predictive reliability.