INTRODUCTION: MRI and MRS are the modalities of choice for the initial diagnosis of brain tumors. Despite the overwhelming amount of performed studies on the available MRI/MRS parameters, the differential diagnosis for individual patients remains difficult: statistical significant differences between patient groups suffering from different types of brain tumors could be shown, but many of these differences are not relevant for individual diagnostics. It is hypothesized that the full potential of MRI/MRS is currently not used in routine diagnostics. To illustrate this claim, this paper will introduce a novel computer aided diagnostic method. It is shown that apparent diffusion coefficients (ADCs) only are sufficient for reliable individual diagnostics of high grade (HGG) versus low grade glioma (LGG) without contrast agents; a task that could not be performed by visual ADC-map inspection alone. METHODS: The key characteristics of the method are: (a.) the determination of so-called histogram- and co-occurrence matrix-based texture parameters (TPMs) from regions of interest drawn by a neuroradiologist encircling the complete tumor-affected volume (high cellular density compartments, edema, and necrosis); (b.) the mean and higher order statistical moments of these texture parameters are used; (c.) these texture features are the input parameters to a random forest classification algorithm (RFC) which is trained to perform the classification, i.e. the diagnosis. A random forest is a classification method consisting of a combined ensemble of decision trees. It is a highly flexible statistical model allowing capturing multiple interactions and non-linearities without over-fitting the data, yielding high accuracy predictions. Moreover, the RFC can be utilized for feature selection identifying a subset of important feature within up to thousands of features. The performance of the RFC is evaluated by its prediction power on new cases by comparison with the experts' diagnoses. RESULTS: The TPMs of a total of 184 ADC-maps were extracted from 18 grade-II and 21 grade-IV glioma patients and were contoured as described above and were classified with the RFC-algorithm. Using the-leave-one out method for validation (on patient level), resulted in 91 out of 104 correctly classified grade-IV, and 61 out of 80 resulted correctly classified grade-II ADC-maps, meaning a classification error of just 17.4% for discriminating the two diseases. CONCLUSION: Using RFC applied on texture parameters of ADC-maps, HGG can automatically be discriminated from LGG with a high level of accuracy, without the need of any MR-contrast agents. Since the method is very simple to apply, and non time consuming its application in a clinical setting is easily feasible. Currently new classifiers on other image modalities than ADC-maps are constructed to discriminate malignant glioma from other malignant brain tumor types.