Assessing and mapping language, attention and executive multidimensional deficits in stroke aphasia

Abstract

There is growing awareness that aphasia following a stroke can include deficits in other cognitive functions and that these are predictive of certain aspects of language function, recovery and rehabilitation. However, data on attentional and executive (dys)functions in individuals with stroke aphasia are still scarce and the relationship to underlying lesions is rarely explored. Accordingly in this investigation, an extensive selection of standardized non-verbal neuropsychological tests was administered to 38 individuals with chronic post-stroke aphasia, in addition to detailed language testing and MRI. To establish the core components underlying the variable patients' performance, behavioural data were explored with rotated principal component analyses, first separately for the non-verbal and language tests, then in a combined analysis including all tests. Three orthogonal components for the non-verbal tests were extracted, which were interpreted as shift-update, inhibit-generate and speed. Three components were also extracted for the language tests, representing phonology, semantics and speech quanta. Individual continuous scores on each component were then included in a voxel-based correlational methodology analysis, yielding significant clusters for all components. The shift-update component was associated with a posterior left temporo-occipital and bilateral medial parietal cluster, the inhibit-generate component was mainly associated with left frontal and bilateral medial frontal regions, and the speed component with several small right-sided fronto-parieto-occipital clusters. Two complementary multivariate brain-behaviour mapping methods were also used, which showed converging results. Together the results suggest that a range of brain regions are involved in attention and executive functioning, and that these non-language domains play a role in the abilities of patients with chronic aphasia. In conclusion, our findings confirm and extend our understanding of the multidimensionality of stroke aphasia, emphasize the importance of assessing non-verbal cognition in this patient group and provide directions for future research and clinical practice. We also briefly compare and discuss univariate and multivariate methods for brain-behaviour mapping.

Keywords: aphasia; attention; executive functions; principal components; univariate and multivariate brain-behaviour mapping.

Figure 1

Overlap of the 38 patients’ lesions.

Figure 2

Percentage of participants with impaired performance on each measure of the non-verbal tests (left) and language tests (right).

Figure 3

Patients’ overall impairment in the non-verbal versus language tests. The percentages of impaired scores correlated significantly (r_s = 0.591, P < 0.01, n = 38, also if patient characteristics were accounted for by means of partial correlations). Symbols and colours denote an individual’s aphasia type based on the BDAE (triangles for non-fluent, circles for fluent patients, for colours see top left legend). More saturated or differently coloured symbols denote two patients in the same spot.

Figure 4

Component loadings and structural correlates associated with each component. (A) The darker coloured bars (from left to right: blue, green, purple, orange, red, pink) represent the loadings on the six components from the combined PCA. The lighter coloured bars represent the loadings on the three components in the separate non-verbal-only PCA (first three columns) and the language-only PCA (last three columns). Loadings < 0.1 are not depicted. MLU = mean length of utterance; WPM = words per minute. (B) Structural correlates associated with each component from the combined PCA. Clusters shown in blue-green were obtained by applying a voxel-level threshold of P ≤ 0.01, clusters in red-yellow correspond to a voxel-level threshold of P ≤ 0.001. A family-wise error correction of P ≤ 0.05 was applied to all clusters. The respective coordinates in MNI-space are indicated on the left side. Figures are in neurological convention (left is left).

Figure 5

Comparison of brain-behaviour mapping results based on the four different methodological approaches. The significant VBCM clusters are shown in blue (voxel-level threshold 0.01) and green (voxel-level threshold 0.001), a family-wise error correction of P ≤ 0.05 was applied to all clusters, and images are thresholded at the respective minimum t-value. The PRoNTo results depict the weights for the winning model if significant (see text), either including the whole brain space or restricting it to lesion territory (n > 3). They are thresholded from −0.005 to −0.0001 (green-blue) and 0.0001 to 0.005 (red-yellow). The negative weights are considered as more meaningful in this approach. The SVR-LSM images show voxels with significant beta weights after permutation testing (n = 10 000, voxel-wise P < 0.005 and cluster-wise P < 0.05). MNI coordinates of slices, from left to right, are z = −25, −10, 5, 20, 35, 50 and they are in neurological convention (left is left). A grey surface indicates that no significant results were found for the respective component and methodological approach.