Susceptibility Induced Loss of Signal:Comparing PET and fMRI

on a Semantic Task

Devlin et al. (in press)

Introduction

• fMRI and PET are both popular techniques for investigating neural correlates of cognitive processes

• Several advantages to fMRI:

• greater temporal and spatial resolution allowing for event-based and trial-based experiments

• doesn’t require exposure to radioactive isotopes

Introduction

• PET findings, however, are not always able to be replicated with fMRI paradigms• Imaging tissue near air-filled sinuses can result

in geometric distortion or worse, loss of BOLD

signal (i.e., susceptibility artifacts)

Courtesy of Dr. James Danckert

Introduction

• Given the close proximity of major language areas to air-filled sinuses, fMRI investigation of language processing has been particularly problematic

• Similar investigations of lexical-semantic processing show activation of these temporal language regions with PET but not with fMRI (Perani et al., 1999; Kiehl et al., 1999)

How are these susceptibility artifacts overcome?

• define regions of interest (ROI) a priori to increase statistical power

• use Worsley et al.’s (1996) small volume statistical correction

Study Objective

• Investigate the usefulness of the Worsley et al.‘s (1996) statistical correction by examining the activation patterns present during a semantic categorizing task using both PET and fMRI

Methods. Semantic Categorization Task

1) Semantic Categorization Task. Subjects read 3 cued words consecutively presented, then made a decision as to whether the fourth (target) word belonged to that category

Example: dolphin, seal, walrus, OTTER

Stimuli: cued and target words displayed for 200ms at 400ms intervals

Responses: “Same” and “Different” were indicated by right- or left-mouse clicks; 1750ms provided after target word for response

2) Letter Categorization Task. Same stimulus and response characteristics but with no lexical or semantic component.

Example: fffffff, ffff, ffffff, FFFFF

Methods. Letter Categorization Task

PET Experiment

Participants. 8 healthy, English speaking males aged 21 – 47 (mean 28)

Stimuli presentation.• Blocked design• Twelve 90 sec. scans (8 semantic

categorization, 4 letter categorization) were presented. 45 sec. of stimuli presentation, 45 sec. of blank screen

• Each subject saw the conditions in a different order

Methods. PET

Functional Imaging Details

• GE Advance PET Scanner

• 35 image planes, each 4.25mm thick.

• Axial field-of-view (FOV) = 15.3cm

• Voxel size = 2.34mm x 2.34mm x 4.25mm

Methods. PET

Results: PET

R. cerebellum

Figure 1. Areas of activation in the Semantic Minus Letter categorization comparison

Inferior temp. gyrus (BA 20)

Ant-medial temp. pole (BA 38)

Broca’s area (BA44/45)

Results: PET

Areas of Activations: R. cerebellum, L. inferior temporal gyrus, L. anterior medial temporal pole, and Broca’s area

• All activations were significant at the cluster level

• All activations except that observed in the temporal pole was significant at the voxel level

Conclusions. PET

• Activation areas are consistent with previous studies looking at lexico-semantic activation

• Both activated regions of the temporal lobe are areas affected by susceptibility artifacts

fMRI Experiment

Methods. fMRI

• Methodology was the same as PET study but with 192 semantic trials (8 trials/block with 12 semantic semantic blocks/session with two sessions)

Methods. fMRI

Biophysical Parameters:

• Varian-Siemens 3T MR scanner used

• A head gradient coil used along with a birdcage head radio-frequency coil

• A gradient-echo EPI sequence was used for image collection (TR 3s, TE 30ms, 64 x 64 resolution, 256mm x 256mm FOV)

• 21 slices with 6mm slice thickness and in-plane resolution of 4mm.

Results. fMRI

R. cerebellum

Figure 1. Areas of activation in the Semantic Minus Letter categorization comparison

Broca’s area (BA 44/45)

BA 8

BA44/45

R-Hem “Broca’s”

Results. fMRI

Areas of activation: L. frontal lobe extending from the inferior frontal gyrus (BA47) into Broca’s area (BA44/45), R. frontal (BA44/45), and L. medial surface of the superior frontal cortex (BA8)

• No reliable activations in the temporal lobe

Applying Statistical Correction to fMRI data• Worsley et al.’s (1996) small volume correction calculation was applied

• The correction showed a reliable region of activation in the L. inferior temporal lobe, a region that was shown to be activated in the PET experiment

• The correction, however, did not reveal activation in the L. anterior-medial temporal cortex, another area that was activated in the PET experiment

Results. fMRI and PET activation overlap

Figure 3. A comparison between the semantic activation in the PET and fMRI experiments. Overlapping activation is shown in yellow. Red = PET; Green = fMRI

Author’s Discussion

• Same task yielded differences in activation in PET and fMRI paradigms

• The statistical correction compensated for the susceptibility artifact in affected temporal regions only when signal loss was relatively small

• Other compensatory measures such as tailored RF pulse sequences may be more successful in recovering signal loss in susceptible temporal regions

Were the fMRI and PET Experiments Exactly the Same?

• Different number of trials (96 vs. 192)

• Length of stimulus blocks (45s vs. 30s)

• Normalization of the images (PET template vs. EPI template)

• Specification of the GLM (diff. in deg. of freedom)

• Different participants

Appropriate Control Condition?

• Letter and Semantic Categorization tasks differed by more than one factor…

• differed in required semantic processing

• differed in whether or not the participant was required to read

Areas of Activation Observed only in fMRI

• Medial frontal (premotor area)