Predictive brain mapping in large populations via multiple modality matrix/tensor factorization
Abstract
Recent innovation from the machine learning community in constructing novel deep learning architectures has demonstrated significant potential in advancing a variety of fields. These applications range from the well-utilized image classification and segmentation architectures to more novel prediction and dimensionality-reduction possibilities. With respect to the latter, the multi-modal and large-dimensionality aspect of neuroimaging applications holds particular promise but have yet to be developed and deployed at scales necessary for such applications as addressing traumatic brain injury research in military service personnel. To address such needs, and building on our previous work involving large-scale neuroimage analysis, we propose the construction of both the theoretical framework and software implementation to comprise a technological platform capable of exploiting prior knowledge for scientific insight. Specifically, we propose a new unifying framework for matrix and tensor-based dimensionality reduction methods, which significantly facilitate the analysis of the various complex neuroscience data types. This framework extends across applications that leverage multiple data streams to enhance both multivariate and longitudinal prediction. This framework will be evaluated on multiple data types and, ultimately, made available through the publicly available BRAIN (Brain Research and Innovation Network) Commons. This framework will generate crucial knowledge for understanding the progression of neurological disease and will be used to identify key imaging (and other) biomarkers predictive of diagnosis, prognosis, and response to therapy.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 26, 2018
- Source ID
- N000141812440
Entities
People
- Magali Haas
Organizations
- Cohen Veterans Bioscience
- Office of Naval Research
- United States Navy