Less is more: neural mechanisms underlying anomia treatment in chronic aphasic patients

See Thompson and Woollams (doi:10.1093/brain/awx264) for a scientific commentary on this article. Naming in aphasia can be facilitated by phonemic cueing, both as an immediate retrieval strategy and as a long-term treatment. Nardo et al. report that immediate and long-term facilitation of naming rely on common neural networks, while whole words and partial cues promote naming via different networks.

The control cues were broadband stimuli derived from the natural single-word speech recordings that were then spectrally rotated and noise-vocoded to remove any phonemic and semantic information. Subjects reported they sounded like words but of a foreign language.
To generate the auditory cues, each target name was digitally recorded (at 44.1 kHz) from a male native English speaker in a soundproof room. Word cues were the spoken whole word tokens. To generate the initial and final phoneme cues each whole word token was cropped at either the offset of the vowel to form the initial cue (e.g. /ka/), or the onset of the vowel to form the final cue (/at/). Initial and final cues were then matched for total auditory duration.
To generate the control cues, each spoken word cue was spectrally rotated (Blesser, 1972), and then submitted to a noise-vocoding routine (Shannon et al., 1995) using a single level of filter band noise vocoding. This procedure leaves the temporal envelop of the spoken token unaltered and preserves the spectro-temporal complexity, while rendering the auditory signal unintelligible by inverting the frequency spectrum.

Potential bias between pools of stimuli at T1
In order to rule out the presence of any (accidental) potential bias between the two pools of stimuli before the therapy, we computed two-tailed paired t-tests between 'to-be-treated' and 'untreated items' at T1. In Experiment 1, the comparisons showed no significant difference for both accuracy (t=0.45, P =0.660) and RT (t=1.00, P =0.331). In Experiment 2, the comparisons showed indeed a significant difference for accuracy (t=-2.56, P =0.021), but not for RT (t=-1.70, P =0.108). Importantly, it should be noticed that the direction of the significant comparison is opposite to our predictions and the effects found at T2 (i.e. TRE>UNT). This means that at T1 'to-be-treated' items were named with a lower accuracy as compared to 'untreated items' (i.e. UNT>TRE).

Procedures
Stimulus-Onset-Asynchrony between pictures and auditory cues 2 Auditory cues were presented simultaneously with each picture (Stimulus-Onset-Asynchrony -SOA=0ms). The time at which the phonemic cue is delivered has important consequences for priming speech production. Previous studies have found robust phonological priming effects on naming with a SOA of 0ms (Schriefers et al., 1990;Meyer and Schriefers, 1991;Starreveld, 2000;de Zubicaray and McMahon, 2009), whereas priming effects decrease when auditory cues precede visual stimuli (Abel et al., 2009).

functional MRI experimental protocol
Each visual stimulus was displayed for 2500ms, preceded by a 1000ms fixation cross and followed by a blank screen for 420ms. Trials were presented in mini-blocks of six stimuli, separated by fixation-only rest periods of 7840ms in order to optimize the BOLD response (Henson, 2006). The inter-trial interval was set to 3920ms to jitter the onset on each trial across acquired brain volumes and vary the spatial acquisition of the functional MRI data.
Overt spoken responses were recorded in the scanner using a dual-channel, noise-cancelling fibre optical microphone system (FOMRI III, http://www.optoacoustics.com). Auditory cues were delivered via an MR-compatible set of headphones (MR Confon, Magdeburg, Germany; www.mr-confon.de).

Computation of Cohen's d
The calculation being, mean naming performance change over time for treated items (experimental condition) minus mean performance change over time for untreated items (control condition), (in case of standardized effect size) divided by the standard deviation of performance change over time for untreated items (control condition). For example, the standardized effect size for naming treatment change between T1 and T2 was calculated as

Preprocessing of functional MRI data
All volumes of interest from each patient were realigned and unwarped, using session and subject-specific voxel displacement maps (Hutton et al., 2002). The functional images were then spatially normalized. This step included skull-stripping using BET running under FSL 3.2.0 (http://www.fmrib.ox.ac.uk/fsl), and a subsequent estimation of segmentation parameters using a unified segmentation routine implemented in the Automatic Lesion 3 Identification (ALI) toolbox (Seghier et al., 2008). Finally, functional data were spatially smoothed with an 8mm FWHM isotropic Gaussian kernel to allow for residual variability after spatial normalization and the application of Gaussian Random Field Theory for corrected statistical inference. To remove low-frequency drifts, the data were high-pass filtered using a set of discrete cosine functions with a cut-off period of 128s.   Table 3 -Correlations between indexes of treatment outcome and demographic/clinical data. Indexes of treatment outcome have been computed as differences between treated items at different time points (as specified in the columns). Legend: RT = reaction time; T1 = pre-treatment measure; T2 = post-treatment measure; T3 = follow-up (at three months); r = Pearson's correlation coefficient; P = associated P -value.