Analysis of breast mass characteristics via ultrasound images plays an important role in cancer screening and treatment planning. Currently, Radiomics is utilized as a useful quantitative tool for medical image analysis. In this work, there are a total of 96 Radiomics features of biological characteristics and image textures from 3 main groups were extracted to analyze. The values and importance values of features and feature groups were quantified by statistical methods. Based on these results, the differences in tumor characteristics and image textures between benign and malignant breast masses were analyzed in depth, which can reveal the biological properties discrimination among groups. Moreover, several classification models, such as the Support Vector Machine and Ensemble learning model, and an oversampling method were utilized to confirm the discriminant performance via Radiomics-based features in the limited-amount dataset. The statistical feature analysis results extensively reveal the biological characteristics and texture differences between benign and malignant tumors. This result orient the further feature extraction process, in order to improve the classification and biological analysis of tumors. In addition, the classification results of the ensemble learning approaches with oversampling technique presented the highest performance, with the F1 score of 87% for benign tumors and 65% for malignant tumors, although training in a small number of images.