It is the goal of this project to develop and use a framework of artificial neural networks coupled with application-specific image processing operators for the analysis and interpretation of clinical, biochemical, physiological, anatomical and cognitive types of data. Recent progress in the use of Optimal Interpolative (OI) and Optimal Multilayer Neural Interpolative (OMNI) nets has demonstrated their special capabilities for machine interpretation of complex patterns in data as well as for intelligent control. These capabilities will be utilized to achieve the stated project goal. Specifically, this project will execute the following tasks: First, our current experience in the neural-network-based analysis of Single-Photon-Emission-Computed-Tomography (SPECT) image data will be ported to the Magnetic Resonance Imaging (MRI) arena for accurate classification of dementia. OI and OMNI neural networks will be especially designed so that, when trained with the multimodality MRI data, they will classify Alzheimer's and other types of dementia with high accuracy even if the set of training samples available is small. To this end, the highly nonlinear separation surfaces (decision boundaries) in the decision space, which these nets are capable of generating, will be used to advantage. Second, the various imaging, cognitive, and the other clinical tests mentioned above will be used to design OI and OMNI neural networks which will discover without supervision (i.e. without labeled samples) meaningful subtypes within a given class of dementia. The underlying algorithms will be based on cluster-oriented concepts. The subtyping will be used to guide future intervention strategies. Third, and finally, in cooperation with the Imaging Core, techniques will be developed under this project for preprocessing and fusing image data modalities which will be used in the analysis by the neural networks mentioned above.
Showing the most recent 10 out of 590 publications
