Topological data analysis (TDA) and topological representation are emerging research directions in medical image analysis, aimed at mining the spatial topological features of diseases such as tumors, exploring biological complex patterns, and constructing predictive tools for clinical diagnosis and treatment. Ultrasound examinations are widely used in the preliminary diagnosis of various tumor diseases due to their convenience, real-time imaging, and non-radiation. However, since ultrasound examinations rely heavily on the subjective experience of doctors and often require pathological gold standards for tumor typing determination, there is an urgent need to develop novel quantitative analysis methods to build the correlation between ultrasound quantitative features and histological information, thereby optimizing clinical decision-making paths. The extensive evaluation and validation studies of TDA in ultrasound imaging are still scarcely reported.We proposed a novel ultrasound topological representation method, termed WT-TDA, which is based on wavelet transformation to enhance topological feature representation and combines SHAP-based feature selection to optimize modeling effects. We evaluated the algorithm on three ultrasound image datasets to verify the potential of topological analysis in clinical diagnosis.The WT-TDA demonstrated ideal tumor diagnosis performance across breast, thyroid, and kidney ultrasound datasets, achieving the test accuracy of 0.932, 0.805, and 0.888 and test AUCs of 0.915, 0.805, and 0.889, respectively. Additionally, WT-TDA enabled the extraction of a set of ultrasound topological features that are beneficial for clinical analysis, and SHAP analysis enhanced the interpretability of the topological models.The study verifies the persistent homology of ultrasound images and demonstrates the potential application of WT-TDA in the benign and malignant diagnosis of ultrasound tumors, which can help optimize ultrasound diagnosis and provide decision support for ultrasound doctors.Copyright © 2025 Elsevier B.V. All rights reserved.