Objective:To assess the feasibility and accuracy of a deep learning model based on two-dimensional ultrasound images for preoperative prediction of lymphovascular invasion (LVI) in breast cancer, and compare its results with traditional interpretations by ultrasound physicians, providing novel imaging insights to support personalized treatment decisions and precision medicine.MethodsA retrospective analysis was conducted on patients diagnosed with breast cancer through postoperative pathology, who underwent breast surgery in Sixth People's Hospital Affiliated to Medical College of Shanghai Jiao Tong University between January 2020 and December 2023. Standardized grayscale ultrasound images of lesions were processed, and features were extracted using a convolutional neural network (CNN) -based deep learning model to predict LVI. The model's performance was evaluated using accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC). Comparisons were made with the diagnostic performance of ultrasound physicians with varying levels of experience, and the impact of the deep learning model on improving physician diagnostics was analyzed.ResultsA total of 232 patients were included, with a total of 232 lesions. The dataset was divided into a training set (185 cases) and a validation set (47 cases) in an 8∶2 ratio. Among the 232 patients, 102 cases (43.97%) were confirmed to have LVI by postoperative pathology. The deep learning model (support vector machine) achieved AUCs of 0.917 (95% CI 0.877-0.956) and 0.863 (95% CI 0.750-0.975) in the training and validation sets, respectively. For the validation set, the accuracy, sensitivity, and specificity were 83.0%, 85.7%, and 80.8%, respectively. In comparison, ultrasound physicians with 5-10 years and 10-15 years of experience achieved AUCs of 0.623 and 0.709, respectively, which were significantly lower than the deep learning model (P<0.05). When assisted by the deep learning model, the AUC for the physician with 10-15 years of experience increased to 0.914, with corresponding improvements in accuracy (91.5%), sensitivity (90.5%), and specificity (92.3%).ConclusionThe deep learning model based on two-dimensional ultrasound images demonstrated superior accuracy and reliability in the preoperative prediction of LVI in breast cancer. It significantly outperformed traditional diagnostic approaches by ultrasound physicians and showed potential as a clinical auxiliary tool to improve the precision of preoperative assessments of LVI, supporting personalized patient treatment planning.