Enhanced differentiation of breast lesions through integration of microvascular flow imaging and machine learning algorithms.
Authors
Affiliations (2)
Affiliations (2)
- Department of Ultrasound, Hangzhou Women's Hospital (Hangzhou Maternal and Child Health Care Hospital), Hangzhou, Zhejiang, China.
- Department of Ultrasound, Affiliated Hangzhou First People's Hospital, School of Medicine, Westlake University, Hangzhou, Zhejiang, China.
Abstract
Breast cancer diagnosis relies on imaging, yet conventional Doppler ultrasound possesses limitations in visualizing tumor microvasculature. This study aimed to compare Microvascular Flow imaging (MV-Flow) with Color Doppler Flow Imaging (CDFI) and evaluate a machine learning (ML) framework integrating MV-Flow parameters for differentiating breast lesions. In this prospective study, 101 breast lesions from 79 patients underwent grayscale ultrasound, CDFI, and MV-Flow. Adler grading and the quantitative Vascular Index (VI) were obtained. Based on six features (age, size, CDFI/MV-Flow Adler grades, VI, BI-RADS), eight ML models were trained (80% data) and tested (20% data). Model performance was evaluated, and SHAP analysis identified key predictors. MV-Flow demonstrated superior sensitivity, detecting blood flow in 16 lesions missed by CDFI, and showed significantly higher inter-observer agreement (weighted Kappa=0.68 vs. 0.51 for CDFI). The median Vascular Index (VI) was markedly higher in malignant lesions (20.25) compared to benign ones (3.10, P<0.001). The diagnostic AUC for MV-Flow Adler grade, VI alone, and their combination were 0.874, 0.823, and 0.888, respectively. Among eight machine learning models trained on six clinical and sonographic features, the K-Nearest Neighbors model achieved the best performance on the independent test set with an accuracy of 0.927 and an F1-score of 0.947. SHAP analysis identified BI-RADS category and patient age as the most important predictive features in the model. MV-Flow outperforms CDFI in depicting breast tumor microvasculature. ML models integrating MV-Flow parameters can optimize diagnostic accuracy, offering an objective tool for clinical decision support.