Investigation into resting state connectivity using ICA

1
Investigation into resting state connectivity
using ICA

• By Beckmann, DeLuca, Devlin and Smith
• Appeared in Phil, Trans. R. Soc (2005)
• Presented at fMRI Meeting, 09/20/2006

2
Motive

• In fMRI image, statistical test is based on the
  contrast of measured image intensity with
  recordings obtained at rest
• But the baseline, resting-state BOLD signal is
  not well defined.
• Thus, characterize and identify the possible
  origins of slow variations in BOLD signal at
  resting state
• It is not hypothesis driven experiment,
  nonparametric technique (ICA,PICA) was sought!

3

• ICA a technique to decompose data matrix (timep
  voxelN) into a set of time courses and
  associated spatial maps.
• PICA an extension of ICA
• By assuming that time course data are generated
  from a set of q(ltp) statistically independent
  non-Gaussian sources corrupted by additive
  Gaussian noise

4
Decomposition in ICA
5
Schematic of PICA
6
Estimation overlapping maps using ICA

• Two source signals s1 and s2 with length N
  without Gaussian noise
• The spatial correlation
• Two source signals s1 and s2 with length N with
  Gaussian noise
• The spatial correlation is
• Thus, the noise process will decrease the spatial
  correlation

7
Estimation of largely overlapping signals in the
presence of noise

• Two true signals with spatial correlation of 0.5
• Use regression method, PCA, PICA, and PICA with
  threshold to estimate the true source signals
• Threshold level relates to loss function, i.e.,
  false positive vs. false negative

8

• Contaminated by the Gaussian noise (sigma3)
• Spatial correlations of estimated maps
• Least Square Estimate 0.27
• Principle Components maps 0.0
• Probabilistic ICA 0.0
• Thresholded maps 0.47

9

• Estimated maps and correlations

10
Experiment Design for 4 issues

• Seed-voxel based correlation vs. PICA
• 200 vol. from single subject under rest and
  finger tapping (30s on/off block design)
• 1.5T, TR3s, TE40ms
• Evaluate the extent to which neural effects can
  be distinguished from non-neural physiological
  effects such as aliased cardiac and respiratory
  cycle, 3T and two experiments with TR120ms vs.
  TR3s

11
Experiment Design for 4 issues

• Determine whether low frequency resting
  fluctuation appear within grey matter or are
  driven by contributions from larger blood
  vessels, 300 vol., 3T, TR3s
• Spatial consistency of resting-state patterns
  across subjects, 10 subjects, 200 vol. from each
  subjects, 3T, TR3.4s

12

• Seed-voxel based correlation analysis
• Seed-voxel the coordinate of the highest
  activating voxel
• Based on the seed-voxel, calculate the
  correlation of all time courses against the time
  course of the seed-voxel in order to find a
  temporally consistent resting network
• Limitation the choice of seed-voxel is random
  which can be affected by some noise effects

13

• Seed-voxel is located in the post-central gyrus
  rather than in the mortor cortex and shows
  significant correlation in motor areas
• Two spatial regions (among 40) are identified by
  PICA. Motor area (left) are separated from other
  sources

14
In high temporal sampling (TR120ms), PICA can
separate the cardiac, respiratory and other low
frequency signal fluctuations in resting
state Even in low temporal sampling (TR3s)
which those low frequency signals are aliased,
PICA can separate those sources
15
The resting state fluctuation are well localized
in grey matter and spatially different from Blood
Vessel Network
16
sagital
coronal
axial
Medial visual cortical area
Lateral visual cortical area
Auditory system
Sensory motor system
Visuo-spatial system
Executive control
Dorsal Visual system
Dorsal visual system
Data from 10 subjects were first motion corrected
and coregistered using affine linear registration
into common space of Montreal Neurological Institu
te (MNI) template.