Asymptomatic meningioma growth, which may cause some symptoms, is monitored with follow-up MRI scans. Since that follow-up is not frequent, neurosurgeons require a technique to predict future meningioma growth from follow-up MRI images to reduce the risk of overlooking progression. However, there are few previous studies on predictors and models of meningioma growth, and even appropriate regions of interest (ROIs) have yet to be established. In this study, we proposed a semi-automated MRI-image processing system for exploratory analysis of Radiomics features to find promising growth predictors, while reducing manual operations such as ROI extraction. This system involved four major functions: 1) support for ROI extraction, 2) Radiomics feature calculation and normalization, 3) exploratory t-tests between Radiomics features and tumor growth, and 4) classifier training and its performance test. The prototype system enabled us, for the first time, to perform exploratory t-tests of 107 Radiomics features with four ROIs in the initial follow-up T2 weighted images of 49 cases, and to find 10 features with significant differences between the growth and non-growth groups. These features were also used to train four classifiers (SVM, ridge regression, decision tree, and k-NN), achieving F1 scores of 0.83 in the SVM and ridge regression classifiers. Although the prediction performance is still limited, the potential of this system was strongly suggested in that the classifiers could be trained with significantly fewer cases by using pre-selected Radiomics features as input rather than the images themselves.