We present a computerized image-based method to automatically identify aggressive tumors on combined positron emission tomography and magnetic resonance imaging (PET-MRI) using radiomics texture features from both PET and multi-parametric MRI (MP-MRI). The work aims at investigating the potential use of new composite textures from PET-MRI for the assessment of different biological properties present in cancer and non-cancer regions, and eventually for early detection of malignant tumors in real clinical practice. Towards this goal, a large number of radiomics features are extracted to characterize the intratumoural heterogeneity and microarchitectural morphologic differences within tumors. These image attributes are valuable for determining tumor aggressiveness. The radiomics model was evaluated on three types of cancers (pancreas, gallbladder, and liver). Compared to single image modality (PET or MRI), the fused PET and MP-MRI achieved the best classification performance in differentiating cancer and non-cancer regions with the area of under curve (AUC) of 0.87 for pancreas cancer, 0.89 for gallbladder cancer, and 0.82 for liver cancer. The results indicated that PET-MRI based imaging biomarkers could be useful in identifying aggressive tumors. © Springer International Publishing AG 2017.