Human connectome evaluation to study brain individuality and variability

CESAR F. CAIAFA; FRANCO PESTILLI

Chicago

Conferencia; Neuroscience 2015; 2015

Society for Neuroscience

Diffusion-weighted magnetic resonance imaging (dMRI) and fiber tracking allow mapping the network of brain connections and white matter tracts in living brains; the connectome. By measuring in living brains these technologies allow charting the relation between human behavior, development and aging. Efforts are necessary to improve accuracy in connectome estimation. Recently we developed a method called LiFE (Linear Fascicle Evaluation; Pestilli et al., 2014) for computing connectome accuracy. LiFE computes how well a connectome predicts the measured diffusion signal. LiFE can be used routinely to identify the most accurate connectomes in individual brains. The network properties of the best predicting connectomes can then be studied in individual subjects. The LiFE method is computationally intensive and can be used with large-memory computers more than 50GB of memory. We present a factorized version of the LiFE model that computes connectome accuracy with a small memory footprint ? less than 1 GB. We used several dMRI datasets published by Pestilli et al., 2014 and the Human Connectome Project with 2, 1.5 and 1.25 mm isotropic resolution, 150, 96 and 90 angular resolution and b value = 2000 s/mm2. We used constrained spherical deconvolution and probabilistic tractography (Tournier et al., 2012) to generate candidate connectomes spanning the white-matter of ten human brains. We present a factorized version of the LiFE model. We show that the factorized model has accuracy indistinguishable from the original LiFE model and does not require large computer memory. We built the LiFE model using both the original method presented in Pestilli et al., (2014) and a sparse multiway array decomposition method (Caiafa et al., 2013). For each brain and connectome we show the multiway approach fits the diffusion data as well as the original LiFE model. We use the new factorized LiFE model to replicate several results of the publication (Pestilli et al., 2014). Error in replicating the results is below 2%. The model use less than 1 GB of computer memory.Current best practice in the field is to generate a connectome using a single tractography method and then study the properties of the white matter tracts or of the network of brain connections. We propose making connectomes evaluation a routine process. We introduce a connectome evaluation method that is computationally not intensive and can be used to study large populations of human brains such as those provided by the Human Connectome Consortium. We show that the method is valuable to study the network of brain connections and study individuality and variability of the human brain.