Optimizing S-detect classification accuracy for BI-RADS 4 breast nodules using multimodal ultrasound parameters.

Background: S-detect is a deep learning (DL)-based ultrasound tool that automatically classifies breast nodules on grayscale images; however, its diagnostic specificity for Breast Imaging Reporting and Data System (BI-RADS 4) lesions is only 59.57%. Whether quantitative multimodal ultrasound (MUS) parameters can effectively enhance the performance of this tool remains unclear. This study therefore aimed to improve the diagnostic accuracy of S-detect in distinguishing benign from malignant breast nodules by integrating MUS parameters. Methods: Clinical and ultrasound data of 231 female patients diagnosed with BI-RADS type 4 breast nodules from June 2019 to March 2024 were retrospectively included, and S-detect classification results based on grayscale ultrasound images were obtained. MUS parameters were extracted, including Adler blood flow grading, vascular resistance index (RI), calcification, elasticity score (ES), elastic strain ratio (SR), and vascularity index (VI), among others, and meaningful parameters were analyzed to optimize the diagnosis results of benign and malignant breast nodules classified by S-detect. Sensitivity (SE), specificity (SP), accuracy (ACC), receiver operating characteristic (ROC) curve, and area under the curve (AUC) were used to evaluate the performance of S-detect classification before and after optimization. Results: Malignant nodules showed significantly higher SR [median 3.55 (2.39, 4.95) Conclusions: Combining S-detect with multi-modal ultrasound parameters significantly improves the differential diagnosis accuracy of the four categories of lesions of BI-RADS, and provides a reliable basis for clinical decision-making. However, this study has the limitation of a single-center retrospective design, and future multi-center prospective studies are needed for further verification.