Neural organization of speech production: A lesion-based study of error patterns in connected speech

Brielle C Stark; Alexandra Basilakos; Gregory Hickok; Chris Rorden; Leonardo Bonilha; Julius Fridriksson

To evaluate lesion damage significantly associated with word-finding errors in connected speech and naming, we conducted univariate voxel-based lesion-symptom mapping (VLSM; Bates et al., 2003) . This analysis was conducted on the two task groups: CS and PNT. We used the proportion of each type of paraphasia (neologistic, phonemic, semantically related, unrelated) as the dependent variable. Voxelwise data were analyzed at p<.05 (one-tailed) with 5000 permutations (Winkler et al., 2014) . Cluster thresholding was then employed (z>2.33, p<0.01) and corrected for multiple comparisons using nonparametric permutation (5000 permutations; Winkler et al., 2014) . Specifically, the order of operations for this analysis was as follows: 1) compute voxelwise statistics for 5000 permutations, 2) zero all voxels less than z=1.645 (p>0.05), 3) measure the number of voxels in the largest surviving cluster and save this variable, 4) rank order the 5000 observations of max cluster size, 5) threshold at the 100 th largest cluster size (i.e. given random data, we observed only a single cluster of this size in 1% of the permutations) and 6) zero all voxels less than z=1.645 and zero all clusters smaller than the cluster identified in step #5. This cluster thresholding is part of the VLSM software detailed in Wilson et al., (2010) as well as in our own program, NiiStat (https://www.nitrc.org/projects/niistat/). VLSM analyses were conducted on voxels in which at least 10% of participants shared a lesion. To correct for lesion volume, we used Freedman-Lane permutation correction with lesion volume as a nuisance variable (Winkler et al., 2014) . As some paraphasias significantly correlated with apraxia of speech (AOS) severity in the connected speech condition, we also regessed AOS severity (a score of 0 to 4 where 0 means no AOS; scores derived from the Apraxia of Speech Rating Scale, ASRS Strand et al., 2014) from these error types using Freedman-Lane permutation.