In recent years, human brain networks have received great attention since they describe comprehensive maps of structural and functional connections in the brain in relation to cognition, neuropsychiatric and genetics. Existing methods for network analysis partition the brain into a few hundred regions. Functional or structural information is then overlaid on top of the parcellation for further analysis. However, the parcellation of a few hundred regions cannot fully characterize potential differences in the brain anatomy and function among individuals. This project will tackle the challenge by developing computationally efficient mathematical models for building continuous brain networks. We will demonstrate the various uses of the new models including the prediction of individual cognitive abilities. The project will produce new algorithms in network models, deep learning and accompanying codes and processed data that will serve as a testbed for the development of more advanced methods. The impact of the project goes beyond the intended applications and will support more advanced methods in other areas. The project has great potential to reshape the research on how networks are constructed and analyzed. We expect that the continuous brain networks characterize the fundamental nature of individual brains and improve the predictive power to individual differences in terms of cognitive abilities. The project will also provide versatile an open-source toolbox of algorithms for modeling and visualizing large-scale functional and structural brain networks continuously.
Researchers who use existing brain parcellations for building and analyzing brain network models face several challenges: 1) the inherent limitations of using predetermined parcellations for understanding brain organizations in multiple spatial scales; 2) conflicting network topology over the choice of parcellation; 3) decreased sensitivity over multimodal integration. These have been raised as major challenges for the connectome-based prediction of individual cognitive abilities. The prediction models may not perform optimally if the boundary of the brain parcels does not fit the data well. Further, the specific choice of brain parcellations may bias prediction outcomes. Given these limitations, the main goal of the project is to develop computationally efficient mathematical models for building continuous functional and structural brain networks without using existing brain parcellations. Using these novel network constructions, we will develop new computationally efficient deep learning approaches that incorporate the proposed network geometry and predict individual cognitive abilities without relying on predefined parcellations. We will demonstrate the use of the continuous networks to understand brain organizations in multiscale levels and predict individual cognitive abilities such as intelligence, working memory, attention and cognitive controls.
This award is being co-funded by the CISE Information and Intelligent Systems (IIS) through the CRCNA and BRAIN Programs, and the MPS Division of Mathematical Sciences (DMS) through the Mathematical Biology Program.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.