Opportunities and challenges in application of VASARI features in neuro-oncology

The application of objective radiological data in glioma research remains an underexplored opportunity in neuro-oncology. Despite multiple efforts to incorporate radiomic features in glioma analysis, the lack of a unified imaging reporting system in gliomas prevents large-scale, multi-center scientific collaborations from integrating comparable imaging variables to their work. VASARI lexicon, a set of 30 semantic radiomic features using controlled vocabulary and visual guides, was created to become a shared language for neuroscientists evaluating gliomas. It has been proven, that VASARI features are not only reliably correlated with various genetic aberrations but are also increasingly included in numerous statistical and machine learning models for predicting overall survival, tumor recurrence, glioma grade, and tumor molecular characteristics. In this review, we present an overview of VASARI application in prediction modeling, including the best-performing models and relevance of VASARI features depending on the predicted outcomes. We delve into the opportunities related to the inclusion of VASARI along with clinical, computational radiomic, and multi-omic features to construct combined neuro-oncological prediction models. Lastly, we discuss the challenges and limitations of currently performed models as well as the methods to overcome them using automatic feature extraction solutions.