The solitary pulmonary lesion (SPL) is one of the most common imaging findings. In this study, lung MR using T2 weighted imaging were acquired and analyzed using radiomics. Both 2D and 3D features combined with machine learning approach were compared to determinate an optimal model for differentiation of SPLs. We found that the radiomics signatures derived from 3D features outperformed that from 2D features. The best 3D radiomics model for the classification was a combination of principal component analysis(PCA), analysis of variance(ANOVA) and linear discriminant analysis (LDA). The T2WI-based radiomics model shows potential in differentiating malignancy from benign SPLs.