Capturing multidimensionality in stroke aphasia: mapping principal behavioural components to neural structures

Abstract

Stroke aphasia is a multidimensional disorder in which patient profiles reflect variation along multiple behavioural continua. We present a novel approach to separating the principal aspects of chronic aphasic performance and isolating their neural bases. Principal components analysis was used to extract core factors underlying performance of 31 participants with chronic stroke aphasia on a large, detailed battery of behavioural assessments. The rotated principle components analysis revealed three key factors, which we labelled as phonology, semantic and executive/cognition on the basis of the common elements in the tests that loaded most strongly on each component. The phonology factor explained the most variance, followed by the semantic factor and then the executive-cognition factor. The use of principle components analysis rendered participants' scores on these three factors orthogonal and therefore ideal for use as simultaneous continuous predictors in a voxel-based correlational methodology analysis of high resolution structural scans. Phonological processing ability was uniquely related to left posterior perisylvian regions including Heschl's gyrus, posterior middle and superior temporal gyri and superior temporal sulcus, as well as the white matter underlying the posterior superior temporal gyrus. The semantic factor was uniquely related to left anterior middle temporal gyrus and the underlying temporal stem. The executive-cognition factor was not correlated selectively with the structural integrity of any particular region, as might be expected in light of the widely-distributed and multi-functional nature of the regions that support executive functions. The identified phonological and semantic areas align well with those highlighted by other methodologies such as functional neuroimaging and neurostimulation. The use of principle components analysis allowed us to characterize the neural bases of participants' behavioural performance more robustly and selectively than the use of raw assessment scores or diagnostic classifications because principle components analysis extracts statistically unique, orthogonal behavioural components of interest. As such, in addition to improving our understanding of lesion-symptom mapping in stroke aphasia, the same approach could be used to clarify brain-behaviour relationships in other neurological disorders.

Keywords: aphasia; language processing; lesion symptom mapping; phonology; semantics.

Figure 1

Participants’ scores on phonological and semantic factors, split according to scores on the cognitive factor (above versus below the median). (A) Participants with lower scores on the cognitive factor; (B) participants with higher scores on the cognitive factor. Dual colour and letter coding reflects each participant’s aphasia classification (Table 1).

Figure 2

Lesion overlap map showing the distribution of participants’ lesions (n = 31). Lesions were identified using Seghier et al.’s (2008) automated software. Colour scale indicates number of participants with a lesion in that voxel.

Figure 3

Results of a VBM analysis comparing tissue concentration of participants with anomic aphasia (n = 9), cluster size 25 668, Broca’s aphasia (n = 8), cluster size 37 447, or mixed non-fluent aphasia (n = 6), cluster size 41 357, to healthy older controls (n = 19). Image threshold (t) 3.0–6.0. Results are presented at P < 0.001 voxel-level, P < 0.001 FWE-corrected cluster-level. Analyses were not conducted for aphasic subgroups with n < 5 participants (global = 3, transcortical motor aphasia = 1, transcortical sensory aphasia = 1, Wernicke’s = 2, Wernicke’s/conduction = 1).

Figure 4

Results of a VBM analysis comparing tissue concentration of participants with the lowest scores on picture naming (n = 9), cluster size 51 996, delayed non-word repetition (n = 8), cluster size 45 843, or spoken word-to-picture matching (n = 6), cluster size 46 834, to healthy older controls (n = 19). Image threshold (t) 3.0–6.0. Results are presented at P < 0.001 voxel-level, P < 0.001 FWE-corrected cluster-level. Numbers in each group were chosen for comparability with BDAE Subtypes in Fig. 3.

Figure 5

Regions found to relate significantly and uniquely to phonological (A) and semantic (B) performance in VBCM analyses. Hot overlays are clusters significant at P < 0.001 voxel-level, P ≤ 0.001 FWE-corrected cluster-level and which were interpreted in the text. Cluster sizes 2622 (A) and 856 (B) voxels. Image threshold (t) 2.0–6.0. ACTC (blue/green) overlays are clusters significant at P < 0.01 voxel-level, P ≤ 0.001 FWE-corrected. Image threshold (t) 1.0–5.9.

Figure 6

Regions found to relate significantly to delayed non-word repetition (A) and spoken word-to-picture matching (B) performance in VBCM analyses. Hot overlays are clusters significant at P < 0.001 voxel-level, P ≤ 0.001 FWE-corrected cluster-level and which were interpreted in the text. Cluster sizes 926 (A) and 1707 (B) voxels. Image threshold (t) 2.0–6.0. ACTC (blue/green) overlays are clusters significant at P < 0.01 voxel-level, P ≤ 0.001 FWE-corrected. Image threshold (t) 1.0 – 5.9.

Figure 7

Overlap of lesion (green outline) with phonological (hot) and semantic (cool) clusters from the voxel-performance analysis for example Patients AL, DM, KS and LM. Lesion outlines were generated using Seghier et al.’s (2008) automated software (see main text for details). Axes reflect the participants’ scores on phonological and semantic factors of the PCA, as per Fig. 1. Patient AL = A6, Patient DM = B3, Patient LM = G2, and Patient KS = transcortical sensory aphasia in Fig. 1.