VLSM

1
Voxel-based lesionsymptom mapping An
application to speech fluency and language
comprehension Stephen M. Wilson,1 Elizabeth
Bates,2 Ayse Pinar Saygin,2 Frederic Dick,2
Martin I. Sereno,2 Robert T. Knight3,4 Nina F.
Dronkers4,5 1. Neuroscience Interdepartmental
Program, University of California, Los Angeles.
2. Department of Cognitive Science, University of
California, San Diego. 3. Helen Wills
Neuroscience Institute, University of California,
Berkeley. 4. Center for Aphasia and Related
Disorders, VA Northern California Health Care
System. 5. Departments of Neurology and
Linguistics, University of California, Davis.
Results Figure 1. T-maps for
fluency (upper panels) and
comprehension (lower panels). Voxels shown
colored were significantly associated with
behavioral deficits on these measures,
controlling the false discovery rate at p
0.05. Fluency was most affected by lesions in
the insula (fig. 1b) and in the arcuate/superior
longitudinal fasciculus in parietal white matter
(fig. 1c). Auditory comprehension was most
affected by lesions in the middle temporal gyrus
(MTG) (fig. 1d), with significant contributions
also seen in dorsolateral prefrontal cortex (fig.
1e) and parietal association cortex (fig. 1f).
Correlated lesions As is always the case in
lesion studies, an area may emerge as relevant
because it plays a direct causal role, or because
of a diaschitic effect involving highly
correlated lesions some distance away. Could the
insula be emerging just because it is close to
Brocas area? Could the MTG be identified just
because it is adjacent to Wernickes area? VLSM
allows questions such as these to be
addressed. We selected four voxels-of-interest
based on anatomical criteria in the centers of
Brocas area, the anterior insula, Wernickes
area (posterior superior temporal gyrus), and the
MTG. Four maps were then constructed by
performing an ANCOVA at each voxel covarying out
the lesion status (lesioned vs. intact) of the
reference voxel-of-interest (fig.
2). Figure 2. Maps of ANCOVAs
exploring hypotheses about diaschitic effects by
factoring out (a) Brocas area for fluency (b)
the anterior insula for fluency (c) Wernickes
area for comprehension (d) the MTG for
comprehension. These maps confirm that the
anterior insula is more important than Brocas
area for fluency, and the MTG is more important
than Wernickes area for comprehension.
Sample size and replicability How many patients
are needed in a dataset for lesion-symptom
mapping results to be reliable? To answer this
question, we constructed VLSM maps for fluency
based on random subsets of our 101 patients, and
then quantitatively compared maps based on
different subsets of patients. For nine different
sample sizes (10, 15, 20, , 50), we constructed
and compared 10 pairs of maps. For example, to
make a pair of maps for a sample size of 20, we
would (1) select 20 patients at random (2)
construct a map (3) select a different set of 20
patients at random (4) construct another map
(5) compare these two maps. Maps of d
(standardized difference in means) rather than t
were used, to avoid overestimating map
similarities because of consistency in the
distribution of lesions. Maps were compared by
smoothing with a circular filter (radius 15
mm), then fitting a regression line (treating
voxels as subjects) to determine what percentage
of the variance in one map could be explained by
the other map. When replicability
isplotted as a functionof sample size (fig.
5),a positive relationshipcan be observed.
Inpractice, a samplesize of 20 or 30
shouldgenerally prove reliable,though
individualresults may vary! Discussion and
conclusions Methodological VLSM is an improvement
on previous lesion-symptom mapping techniques
because it uses all available information,
eliminating reliance on cutoff scores, clinical
diagnoses, or specified regions of
interest. ANCOVAs can be used to examine effects
based on correlated lesions. Sample sizes of 20
to 30 give reasonably consistent results.
Reliability increases with sample
size. Substantive The anterior insula is more
important for fluency than Brocas area. The MTG
is more reliably associated with comprehension
deficits than Wernickes area. References Bates,
E., Wilson, S. M., Saygin, A. P., Dick, F.,
Sereno, M. I., Knight, R. T. Dronkers, N. F.
(2003). Voxel-based lesion-symptom mapping.
Nature Neuroscience, 6, 448-450. Chao, L. L.
Knight, R. T. (1998). Contribution of human
prefrontal cortex to delay performance. Journal
of Cognitive Neuroscience, 10, 167177. Dronkers,
N. F. (1996). A new brain region for coordinating
speech articulation. Nature, 384,
159161. Kertesz, A. (1979). Aphasia and
associated disorders Taxonomy, localization, and
recovery. New York Grune Stratton. Saygin, A.
P., Wilson, S. M., Dick, F., Dronkers, N. F.
Bates, E. (2003a). Neural correlates of
non-linguistic impairments in auditory and visual
modalities in left-hemisphere damaged patients
revealed with voxel-based lesion-symptom mapping.
HBM 2003. Saygin, A. P., Wilson, S. M., Hagler,
D. Sereno, M. (2003b). Brain areas involved in
the processing of biological motion
Lesion-symptom mapping and fMRI. HBM 2003.
Address for correspondence Stephen Wilson,
Neuroscience Interdepartmental Program, UCLA,
1320 Gonda Center, 695 Young Drive South, Los
Angeles, CA 90095-1761. E-mail
stephenw_at_ucla.edu.
Introduction Until the last few
decades,cognitive neuroscience reliedlargely on
cases which cameto autopsy to provideevidence
bearing on brain/behavior relationships. The
advent of noninvasivestructural imaging (CT,
MRI, etc.) has made possible the acquisition of
much larger datasets. This makes it important to
establish quantitative methods for making
inferences about the relationship between lesions
and the symptoms they produce. Voxel-based
lesion-symptom mapping (VLSM) is an approach to
this goal (Bates, et al., in press). VLSM
exploits continuous behavioral and continuous
lesion information. The method does not require
patients to be grouped either by lesion site or
by arbitrary behavioral cutoffs. Instead,
statistical analyses of the relationship between
tissue damage and behavior are carried out on a
voxel-by-voxel basis, and the resultant
statistics are plotted as color maps which depict
the degree of behavioral involvement for each
voxel. Previous approaches There are two common
approaches to making inferences based on samples
of patients Groups defined by behavior Patients
are divided into groups according to whether or
not they exhibit a particular behavioral deficit
(e.g. apraxia of speech in Dronkers, 1996
aphasic syndromes in Kertesz, 1979). The lesions
of the impaired patients are overlaid to
determine if there is a common locus of
infarction. An overlay of the spared patients
lesions can also be made to confirm that the
identified area is spared. The main limitation of
this approach is that it forces a yes/no decision
to be made about whether a patient is impaired.
In situations where deficits are not binary, an
arbitrary cutoff must be stipulated and
information about varying degrees of performance
is lost. Groups defined by lesions Patients are
divided into groups based on broad lesion
locations (e.g. dorsolateral prefrontal cortex in
Chao Knight, 1998). Behavioral measures of
interest are then compared to other groups or to
controls. This approach therefore dichotomizes
the data not based on behavior but based on
lesion. Again, potentially valuable information
about patients lesion locations cannot enter
into the analysis. Method VLSM is fully
continuous in both the brain and behavior
domains. In this study, we analyzed speech
fluency and language comprehension data for 101
left-hemisphere-damaged stroke patients (details
below). The basic VLSM method is as follows At
each voxel, patients are divided into two groups
according to whether their lesions do or do not
include that voxel. These groups are then
compared (e.g. with a t-test) and the resultant
statistics are displayed as color
maps. Behavioral measures consisted of subscales
from the Western Aphasia Battery (Kertesz, 1979).
Fluency scores reflect a combination of
articulatory, word finding, and sentence
production skills, while the auditory
comprehension measure represents the average
score on yes/no questions, single word
recognition, and enactment of 1, 2 and 3-part
commands. Patients lesions were reconstructed
onto templates by a board-certified neurologist
(RTK) with expertise in behavioral neurology but
blinded to the clinical status of the
patient. Patients were at least one-year post
onset of the stroke at the time of behavioral
testing. All were native English speakers, with
normal or corrected-to-normal vision and hearing.
The study was approved by the VA Northern
California Health Care System and UCSD Human
Research Protection Programs, and all
participants gave informed consent. VLSM
algorithms were programmed in MATLAB (The
Mathworks, Natick, MA) and are freely available
online at http//crl.ucsd.edu/vlsm.
Figure 3. Reasonable replicability with a sample
size of 20. (a) one fluency map (b) the same map
smoothed (c) another fluency map based on
different patients (d) this map smoothed (e)
the correlation between the maps.
Figure 4. Excellent replicability with a sample
size of 50. (a) one fluency map (b) the same map
smoothed (c) another fluency map based on
different patients (d) this map smoothed (e)
the correlation between the maps.
Figure 5. Replicability as a function of sample
size
More applications of VLSM at HBM 2003 Saygin et
al. (2003a) uses VLSM to examine the neural
correlates of non-linguistic impairments in
aphasia (environmental sound recognition and
pantomime interpretation). Poster 1414 Saygin,
et al. (2003b) is a study of biological motion
processing employing both fMRI and lesion-symptom
mapping. Poster 1551