The objective of this project is to develop optimized image analysis pipelines for pharmaceutical evaluation based on functional neuroimaging, in particular based on positron emission tomography (PET). Strength of drug effect will be assessed using statistical measures of detection power, reproducibility of spatial activation patterns, and regional power estimates. Spatial parametric images will be computed for each of the prediction models investigated. A wide range of predictive models, including many that are new to the functional neuroimaging field, will be implemented and evaluated within a sophisticated statistical resampling framework. A grid-aware software package will be developed based on the results obtained. This software is intended to be run in-house in a distributed computing environment at Predictek to provide state-of-the-art optimized image analysis services to its customers. A scaled-down user-friendly Java implementation of some of the best- performing methods (based on evaluations performed in the project) will also be developed for distribution to end users wishing to perform advanced analyses themselves. The proposed project is expected to yield significant research findings, in addition to augmenting Predictek's neuroimage analysis business. ? ? ?
| Wernick, Miles N; Yang, Yongyi; Brankov, Jovan G et al. (2010) Machine Learning in Medical Imaging. IEEE Signal Process Mag 27:25-38 |
