Predicting functional cortical ROIs via joint modeling of anatomical and connectional profiles

Localization of functional cortical ROIs (regions of interests) in structural data such as DTI and T1-weighted MRI has significant importance in basic and clinical neuroscience. However, this problem is challenging due to the lack of quantitative mapping between brain structure and function, which relies on both the availability of benchmark training data such as task-based fMRI and effective machine learning algorithms. By using task-based fMRI derived ROIs as benchmarks, this paper presents a novel approach that develops predictive models of those ROIs based on concurrent DTI and T1-weighted MRI datasets within a machine learning paradigm. Particularly, in application stage, the predictive models are only applied on the structural datasets to predict functional ROI locations, which are evaluated by cross-validation studies, independent tests and reproducibility studies. We envision that these predictive models can be widely applied in scenarios that have only DTI and/or MRI data, but without task-based fMRI data.