Event Abstract Back to Event Using local neural heterogeneity to both predict and track in language recovery Jeremy J. Purcell1*, Robert Wiley1 and Brenda Rapp1 1 Johns Hopkins University, United States Introduction Little is known about the changes in neural representations that support post-stroke recovery. Recent work suggests that the local differentiation of neural responses reflects representational integrity and learning, with differentiation increasing with expertise and learning (Jiang et al., 2013). We utilize a novel technique – Local-Heterogeneity Regression (Local-Hreg) Analysis - to quantify local neural heterogeneity (Purcell et al., under review). We apply this approach to examine neural representations both before and after behavioral training for acquired dysgraphia. We find that within left ventral occipitotemporal cortex (vOTC), local Hreg values prior to training predict response to training and that pre-post training changes in Local-Hreg values in this area index training-based improvements in spelling performance. Methods Participants were 20 individuals with post-stroke, chronic dysgraphia. Individualized word sets were developed and participants were trained for approximately 3 months. All improved to above 90% accuracy levels. FMRI with a spelling task was carried out at pre- and post-training which included spelling with TRAINING words and a non-spelling CONTROL. A whole-brain, Local-Hreg searchlight analysis was performed. For each search space, it employed a general psychophysiological interaction regression-based analysis (gPPI; McLaren et al., 2012) that uses the neural response of the center voxel as the basis for predicting the surrounding voxels. Local-Hreg indexes voxel-to-voxel interactions within each searchlight thereby quantifying the relative similarity/dissimilarity of the task-specific BOLD responses across adjacent voxels: the lower the local cross-voxel interaction, the higher the local heterogeneity. A whole-brain local-Hreg and traditional GLM analyses were carried out comparing the neural responses to the TRAINING condition at pre-training versus post-training time-points. fMRI Results The whole-brain searchlight identified one Hreg cluster in the left vOTC (non-parametric permutation corrected p = 0.12, cluster size = 22 voxels; see Figure 1). As depicted on the right of Figure 1, the next largest cluster had a p-value = 0.4 indicating that this left vOTC cluster is the most relevant cluster in a whole-brain searchlight. The Local-Hreg values from this cluster were used to examine the relationship of Local-Hreg to spelling behavior. 1) Using a linear mixed effects (LME) model including fixed variables (age, lesion volume, severity, deficit type, time post stroke, training duration) and random effects (participant, voxel), we found that local-Hreg values at pre-training significantly predicted the magnitude of training-based behavioral improvement (beta= 0.077, p=0.0000043, df=867). For visualizing this relationship, a scatter plot of the raw data is shown in Figure 2.A. 2) Using the same LME with local-Hreg change as the dependent variable (i.e. Post Training – Pre Training), we found that local-Hreg change was significantly inversely related to the magnitude of performance change on the training items (beta= -0.033, p=0.048, df=1741). For visualizing this relationship, a scatter plot of the raw data is shown in Figure 2.B. Discussion Participants with the highest values of local neural differentiation prior to training were most likely to benefit from training, suggesting that those with the highest “orthographic tuning/integrity” prior to training benefited the most. In addition, participants who improved the most required the least amount of change in the neural differentiation of orthographic representations, presumably because of the relatively high level of integrity of their spelling systems to begin with. This work provides a novel approach for quantifying the local heterogeneity of orthographic representations, and reveals that this measure is can be used to both predict response to treatment and track neural changes in neural differentiation. Figure 1 Figure 2 Acknowledgements Acknowledgements: Jennifer Shea and the multi-site, NIDCD-supported project examining the neurobiology of language recovery in aphasia (DC006740). References