<p>Accurate discrimination between benign and malignant lymph node lesions is essential for guiding clinical decision-making. As a non-invasive, non-ionizing, cost-effective, and real-time imaging technique, ultrasound has become a critical tool for lymph node evaluation. Although integrating multiple ultrasound modalities including B-mode, Doppler, elastographic, and contrast-enhanced ultrasound can improve diagnostic comprehensiveness, acquiring all these modalities in routine practice is often impractical. In this work, we propose a novel diagnostic framework for scenarios with missing modalities, termed the Gradient Boosting Machine-Based Residual Transfer with Privileged Information (GBM+). Our approach treats hard-to-acquire modalities as the privileged information available only during training, while easily accessible modalities acquired during routine ultrasound examinations serve as the standard information used in both training and testing. Unlike conventional boosting algorithms, each iteration in our GBM + not only corrects residual errors but also incorporates insights from privileged features, facilitating progressive knowledge transfer. This enables the model to achieve robust and accurate classification using only routine ultrasound modalities during the testing phase. Experimental results demonstrate the superior performance of GBM+, which achieves an area under the curve (AUC) of 0.890, an accuracy of 0.817, and a precision of 0.901. These findings underscore the value of leveraging privileged information to enhance diagnostic performance. By effectively transferring shared knowledge across both data and models, GBM+ enhances the accuracy of lymph node diagnosis and shows great potential for improving the diagnostic efficacy of routine ultrasound examinations.</p>