Learning hemometabolic maps from PET
From hpcwiki
Contents |
People
- Francisco Gómez
Summary
18F-fluorodeoxyglucose positron emission tomography (FDG-PET) is a well established imaging technique to measure the global metabolic consumption of glucose in the brain. Nevertheless, it remains a mildly invasive approach due to the intravenous injection. Recently, we explored the possibility to construct metabolic activity maps out of functional connectivity maps as extracted from resting state fMRI activity (1). These maps were obtained by applying independent component analysis (ICA) to resting state fMRI and subsequently combining only components of neuronal origin. The obtained maps showed significant correlation with PET data from the same subjects. Our original approach used the square root of the ICA-maps before summing up all neuronal components. This step aims to reduce the spatial sparsity initially requested by the ICA signal decomposition to obtain a least sparse signal as expect for the FDGPET metabolic maps. In this work, we propose to improve this previous work by formulating a generalized combination model, which parameters will be estimated using machine learning from samples of both resting state activity and FDG pet. By using this approach we expect to improve our understanding of the neuronal basis of the resting state activity.
Method overview
Data sources
Results (Expected)
- Method conference. Model proposal.
- Journal article. Complete model and application to clinical data.
