Correlations between Categorical Perception of Sounds and White-matterIntegrity in the Temporal Lobe

Rutvik Desai, Einat Liebenthal, Anjali Desai, Colin Humphries, Brinda Ramachandran, Michael Ellingson, Jeffrey R BinderNeurology Department, Medical College of Wisconsin, Milwaukee, WI, USA

METHODSMETHODSSUBJECTS: N=18, normal hearing, no neurological symptoms.STIMULI: Phonemic (P): 7-token /ba/ to /da/ continuum, cued by the F2 transition. Nonphonemic (NP): Analog continuum created by spectral rotation of F1 in the syllables.TASK: Subjects scanned once (PRE scan), trained, and scanned again (POST scan). Before both scans: 2-alternatives forced-choice ABX discrimination of token pairs 3-5(across category in phonemic continuum), and 1-3 and 5-7 (within-category in the phonemiccontinuum). In PRE and POST scans: Identification of phonemic and nonphonemic continua . Training: three separate days; identification task on P and NP continua till criterion (90%accuracy) is reached on each day + one training session just before the POST scan.

FMRI ACQUISITION & ANALYSIS: GE Excite 3T scanner (GE Healthcare, Milwaukee, WI). Anatomical data: 3-D spoiled gradient-echo sequence. Whole brain, 0.86 x 0.86 x 1 mm3

voxels. Functional data: T2*-weighted, gradient-echo, echo-planar images (TE = 20 ms, flip angle =80°, NEX = 1). TR = 7 s, acquisition time = 2 s, 36 axial slices, 3.44 x 3.44 x 3.50 mm3

voxels. Within-subject analysis: Voxelwise multiple linear regression with reference functionsrepresenting the stimulus conditions (preP, preNP, postP, postNP). General linear tests toobtain contrasts map. Group analysis: Smoothing 4 mm FWHM, threshold at p < 0.02 uncorrected, clusterthreshold 621 µl, corrected p < 0.05 ERP: simultaneous 64-channel recordings (Neuroscan, El Paso, TX) (N = 17).

RESULTSRESULTSINTRODUCTIONINTRODUCTION

SUMMARY AND CONCLUSIONSUMMARY AND CONCLUSION

In a previous fMRI study, we found that an area extending along the leftsuperior temporal sulcus (LSTS) was more responsive to familiar CVsyllables than to comparably complex nonphonemic auditory patterns thatcould not be classified into discrete learned categories (1). In this study, wetest the hypothesis that the increase in activation along the left LSTS duringspeech perception is related to the categorical nature of this process, and notto the linguistic status of phonemes. We train subjects to categorize non-speech sounds to achieve a step-like identification curve. Here, we examinethe correlations between subjects’ ability to perceive novel soundscategorically, and the white-matter (WM) integrity, as measured by DTIscans.

There was a positive correlation between the ability to perceivenovel sounds in a categorical manner without training (preNP CPI),and the white-matter integrity (FA values) near the middle (y = ~-13) and posterior (y = ~-47) LSTS. The R posterior STS (y = ~-50to -57) also showed some correlation.

The improvement in categorical perception of NP sounds aftertraining (postNP CPI - preNP CPI), was also correlated with FA inthe middle (y = ~-16) and posterior (y = ~-44) LSTS.

The amplitude of P220 ERP component was also indicative ofcategorical perception. In both preP and preNP conditions, it wascorrelated with FA in posterior LSTS region (y = ~-40 to ~45) onthe inferior longitudinal fasciculus.

Thus, subjects with higher WM integrity in regions near middleand posterior LSTS appear to have an advantage in perceiving,and learning to perceive, novel sounds in a categorical manner.This is consistent with a role of this region in subserving auditorycategorical perception, and not with a role as a dedicated speech-processing area.1. Liebenthal E. et al., Cerebral Cortex, 15, 1621-1631 (2005).2. Smith, S.M. et al., NeuroImage 23(S1), 208-219 (2004).3. Smith, S.M. et al., NeuroImage, In press.

Download this and related posters at http://www.neuro.mcw.edu/~rhdesaiWork supported by NIDCD Grant R01 DC006287 and NIH M01 RR00058

Stimuli

Pre-training Phonetic

0

20

40

60

80

100

ba2 ba3 ba4 ba5 ba6 ba7 ba8

Tokens

% Id

entif

ied

ba2

500

700

900

1100

RT

(sec

)

Pre-training NonPhonetic

0

20

40

60

80

100

np2 np3 np4 np5 np6 np7 np8

Tokens

% Id

entif

ied

ba2

500

700

900

1100

RT

(sec

)

Post-training Phonetic

0

20

40

60

80

100

ba2 ba3 ba4 ba5 ba6 ba7 ba8

Tokens

% Id

entif

ied

ba2

500

700

900

1100

RT

(sec

)

Post-training NonPhonetic

0

20

40

60

80

100

np2 np3 np4 np5 np6 np7 np8

Tokens

% Id

entif

ied

ba2

500

700

900

1100

RT

(sec

)

BEHAVIORAL

FMRI

Post-Scan

Pre-Scan

CategorizationTraining 1

CategorizationTraining 4

tok2 tok5 tok8

CAT 1

CAT 2

Paradigm

postNP-preNP

DTI ACQUISITION & ANALYSIS: EPI sequence with TE = 82.4 ms, TR = 10.5 s, FOV = 220 mm, matrix 128 x 128 36 axial slices, voxel size 0.859 x 0.859 x 3. 5 mm3

b = 0, 1000 s/mm2 , 25 directions Registration and generation of Fractional Anisotropy (FA) maps with FSL (2), voxelwisestatistical analysis using TBSS (3)

CORRELATIONS WITH FA VALUES

preNP CPI : FA

postNP-preNP CPI : FA

preNP P220 Amplitude : FA

preP P220 Amplitude : FA

CORRELATIONS: Logistic regression was carried out to fit individual identification curves collected during thePre- and Post- scanning sessions A categorical Perception Index (CPI) was defined as the slope of the logistic curve in eachcondition such that higher CPI values indicated a steeper (step-like) identification function FA values for each subject were correlated with CPI on a voxelwise basis Individual FA values were also correlated with the peak amplitude of the P220 ERPcomponent at electrode FC3. All correlation maps thresholded at p < 0.02, clusters smaller than 150 µl removed

Grand Average ERPs (N=17 )Electrode FC3

-10

-5

0

5

-100 0 100 200 300 400 500

Time (ms)

Am

plitu

de (µ

V)

Pre PPre NPPost PPost NP

ERP