Testing the Predictive Power of Structural Neuroimaging in the Estimation of Individuals' Reading and Attentional Abilities

点击翻译按钮获取中文摘要

PROJECT SUMMARY Establishing the incremental predictive validity of neuroimaging is a critical prerequisite for this technology's clinical or educational use outside of medical settings. If specific, well-validated neuroimaging tools provide little to no information above that of clinical interview or neuropsychological assessment, the use and funding of those tools should be more critically evaluated against other methods or funding priorities. If neuroimaging demonstrates unique predictive power for assessment or prediction purposes, however, it may aid in the identification of concerning developmental trajectories, the provision of early intervention, or the prediction of individuals' response to specific interventions. Preliminary data from the applicant's laboratory has demonstrated unique contributions of structural neuroimaging to individual differences in reading and attention using confirmatory structural equation modeling, but these questions have yet to be addressed using a data- driven feature-reduction approach that considers numerous types of demographic, behavioral, and brain- derived measures. This proposal focuses explicitly on structural neuroimaging as it is more easily and more consistently obtained than functional imaging in clinical and research environments, and recent findings indicate that it may even be more highly predictive of behavior than functional imaging. In light of these considerations, this proposal will utilize supervised machine learning to assess the incremental validity of structural neuroimaging above and beyond that of traditional psychological assessment. The goals are this project are to (1) develop and evaluate demographic- and behavior-based predictive models of individuals' reading, inattention, and hyperactivity/impulsivity, (2) replicate and then add neuroanatomical features to these models in order to test structural neuroimaging's incremental predictive validity, and (3) test these models' specificity and discriminant validity for measuring the intended constructs. The long-term goal of this proposal is thus to expand upon the applicant's background in individual difference analyses by developing skills in machine learning so that the incremental validity of multiple neuroimaging modalities can eventually be evaluated. The eventual development of a sufficiently validated predictive model could constitute a behavioral and/or brain-based signature that could serve, along with contextual and functional considerations, as a quantifiable alternative to clinically-based diagnoses. These methods are but first steps toward this distant but worthwhile goal.