Clinical fMRI LEC24 Meta-analysis - Neurometrika · 2001. 10. 17. · Farah & Aguirre (1999) “The...

transcript

Meta-Analysis

Meta-analysis of Single Word Reading (PET)

Meta-analysis and fMRI Comparison

Meta-analysis for literature review

12 years of Single Word Reading Studies

Sources of Localization Variance

•  Resolution limitations of technique •  Different task parameters •  Analysis

– smoothing – motion correction, normalization algorithms – differences in analysis between studies

•  Inter-subject variability

Neuroimaging Meta-analysis

•  Re-analysis of previous findings •  Method of finding consistent activity

within the variance •  Used to answer two questions:

–  Is there a consistent difference between activation patterns for two tasks?

– What brain areas are consistently activated by a task?

Motor Areas of the Medial Wall Picard & Strick (1996)

Ant. Post.

29 studies

Visual Recognition of Words, Objects or Faces

Farah & Aguirre (1999)

“The only generalization that one can make… is that visual recognition is a function of the posterior half of the brain!”

20 studies

Single Word Reading Fiez and Petersen (1999)

Method: •  9 studies of overt single word reading •  Differing control conditions •  147 foci of activation •  Eliminate foci with neighbors (<20mm) from

fewer than half the studies •  Assign remaining foci to 18 clusters •  Find the mean location of each cluster

18 Clusters: L IFG, Motor cortex, SMA, Anterior Cingulate, STG, L Fusiform G, L Lingual G, thalamus, CB

Single Word Reading Fiez and Petersen (1999)

Meta-analysis Design Goals

•  Maximize objectivity – Explicit inclusion criteria – Automated analysis technique

•  Maximize quantification – Unbiased localization – Level of consistency – Statistical analysis of consistency

Hypotheses Which brain areas have been consistently activated during

neuroimaging studies of word reading?

Ho: Activation in previous studies is inconsistent (i.e. random)

H1: Activation in previous studies is consistent (i.e. nonrandom clustering)

Meta-analysis Data

•  Task: Aloud Single Word Reading •  Criteria

– healthy adult subjects – standardized coordinate space – most basal control condition available

•  11 studies •  16 contrasts •  172 foci

Included Studies

Foci of Activation

172 Foci from 11 Studies

Modeling Localization Variance

Gaussian Probability Distribution

FWHM=14.2mm

A 3-D Gaussian at (0,0,0)

Distributions can be summed

Methods

•  Move all foci into the same coordinate space (MNI-space)

•  Divide the space into 2mm cubic voxels •  Model the localization probability

distribution for each focus using a 3-D Gaussian (FWHM = 14.2mm)

•  For each voxel, find the union of the probability of localization for all foci

3-D Probability Density Map

The value at each voxel is the probability that at least one focus in the data-set actually occurs within that voxel.

Max = 1.5%

The Null Hypothesis

•  Ho= Localization of foci is random •  represented by the 3-D probability

density map of a random data set •  1000 random sets of 172 foci were

analyzed •  A histogram of probability values was

created

Probability Values for Ho

Prob = 5.535x10-3 Corrected p = .05

=Ho =Reading

literature

Meta-Analysis Results

corrected p < .05

Turkeltaub et al. Neuroimage 16:765 (2002)

Meta-Analysis Results

Results Summary

Included Studies

Problems Using fMRI in Language Studies

•  Unintended auditory system stimulation

•  Auditory stimulus masking

•  Task interference

•  Susceptibility artifacts from jaw movement

Data Acquisition Concurrent with Task

Data Acquisition

Hemodynamic Response

Hemodynamic Delay and Dispersion

Stimulus

Neuronal Response

Data Acquisition After Task Completion

Data Acquisition

Behavior Interleaved Gradients

Task Epoch Control Epoch

task BIG

Conventional

gradients

fixate read delete

+ rat rat

rat at

fixation reading

+ vocalization

Stimulus

Response

Processes

fixate (2 volumes)

DOG CAP HAT

rest 4 sec

read or delete (5 volumes)

BIG Image Data Acquisition

total of 62 volumes

task 9 sec

words fixation

19-34 group

- 38 - 42 - 46

- 50 - 54 - 58

Sound Deletion

•  15 adult subjects (10 female, 5 male) •  aloud single word reading (high freq., 1 syllable) vs.

fixation •  80 whole-head EPI volumes per condition •  BIG Technique •  Data analysis: head motion correction, global

intensity normalization, high-pass temporal filtering, spatial smoothing, and spatial normalization to MNI-Talairach space

•  Significance threshold set to match meta-analysis for number of significant voxels (Z score > 6.25)

fMRI Methods

Data Acquisition Details

•  BOLD-contrast using EPI

•  1.5 T system •  CP head coil •  TR 13 sec •  TE 40 ms •  3.0 mm3 voxels

fixate read

fixation reading

+ vocalization

Stimulus

Response

Processes

Meta-Analysis vs. fMRI Meta-analysis fMRI

Meta-Analysis vs. fMRI Meta-Analysis fMRI Meta-Analysis fMRI

Central Drift Artifact Deep voxels are surrounded in all directions by potential contributors

Cortical voxels have fewer neighbors: lower likelihood of high probability scores

Deeper voxels =

Higher scores

Central Drift Artifact

Distance Between Loci

Probability value vs. Z score

Values at Closest Peaks

Summary

•  Areas activated during single word reading: –  Primary Motor Cortex –  Superior Temporal Sulcus –  L Fusiform Gyrus –  Cerebellum –  Thalamus –  L Inferior Frontal Gyrus

•  Meta-analysis results were good predictors of location and level of fMRI activation

•  No Angular Gyrus activation

PURE TONES

Clinical fMRI LEC24 Meta-analysis - Neurometrika · 2001. 10. 17. · Farah & Aguirre (1999) “The...

Documents