Functional and structural biomarkers of cognitive outcomes after brain tumor resection

Cognitive explorations have demonstrated the activation of plastic mechanisms in slow-growing brain lesions, generating structural and functional changes. Due to its incidence, it is essential to investigate the reorganization of functional areas in brain tumor patients as well as formulating new approaches for predicting patients quality of life after tumor resection. Following this perspective, we formulated an efficient methodology for postsurgical prognosis prediction, not only in terms of the structural damage but also to measure the neuroplastic changes associated with tumor appearence. Of note, most of previous studies employed a limited number of neuropsychological and clinical features for predicting patient prognosis. Our objective is to optimize the traditional model and to develop a method that can predict outcomes with high accuracy and identify the most significant features for cognitive impairment, working memory, executive control and language outcomes. Our approach is based on the inclusion of a large battery of neuropsychological tests as well as the introduction of grey and white matter morphological measures for model optimization. We employed Support Vector Machine (SVM), Decision Tree, and Naive Bayes algorithms for testing the models and outcomes. Overall, SVM performance showed to be more accurate as compared to Decision Tree and Naive Bayes. Specifically, we found that, by introducing connectivity variables (e.g., grey and white matter measures) Cognitive Status and Working Memory exhibited a predictive improvement. However, Language and Executive Control outcomes were not significantly predicted in none of the models. The importance of the present study resides in the employment of structural and functional variables for postsurgical outcome prediction. We found that connectivity variables are sensitive for predicting the postsurgical quality of life.