A study-specific fMRI normalization approach that operates directly on high resolution functional EPI data at 7 Tesla

Neuroimage. 2014 Oct 15:100:710-4. doi: 10.1016/j.neuroimage.2014.06.045. Epub 2014 Jun 25.

Abstract

Due to the availability of ultra-high field scanners and novel imaging methods, high resolution, whole brain functional MR imaging (fMRI) has become increasingly feasible. However, it is common to use extensive spatial smoothing to account for inter-subject anatomical variation when pooling over subjects. This reduces the spatial details of group level functional activation considerably, even when the original data was acquired with high resolution. In our study we used an accelerated 3D EPI sequence at 7 Tesla to acquire whole brain fMRI data with an isotropic spatial resolution of 1.1mm which shows clear gray/white matter contrast due to the stronger T1 weighting of 3D EPI. To benefit from the high spatial resolution on the group level, we develop a study specific, high resolution anatomical template which is facilitated by the good anatomical contrast that is present in the average functional EPI images. Different template generations with increasing accuracy were created by using a hierarchical linear and stepwise non-linear registration approach. As the template is based on the functional data themselves no additional co-registration step with the usual T1-weighted anatomical data is necessary which eliminates a potential source of misalignment. To test the improvement of functional localization and spatial details we performed a group level analysis of a finger tapping experiment in eight subjects. The most accurate template shows better spatial localization--such as a separation of somatosensory and motor areas and of single digit activation--compared to the simple linear registration. The number of activated voxels is increased by a factor of 1.2, 2.5, and 3.1 for somatosensory, supplementary motor area, and dentate nucleus, respectively, for the functional contrast between left versus right hand. Similarly, the number of activated voxels is increased 1.4- and 2.4-fold for right little versus right index finger and left little versus left index finger, respectively. The Euclidian distance between the activation (center of gravity) of the respective fingers was found to be 13.90 mm using the most accurate template.

Publication types

  • Research Support, N.I.H., Extramural
  • Technical Report

MeSH terms

  • Adult
  • Brain Mapping / instrumentation
  • Brain Mapping / methods*
  • Cerebral Cortex / physiology
  • Data Interpretation, Statistical
  • Echo-Planar Imaging / instrumentation
  • Echo-Planar Imaging / methods
  • Humans
  • Image Processing, Computer-Assisted / methods
  • Magnetic Resonance Imaging / instrumentation
  • Magnetic Resonance Imaging / methods*
  • Psychomotor Performance / physiology