EEG-based neural tracking of linguistic speech representations in people with post-stroke aphasia
Jill Kries1, Pieter De Clercq1, Marlies Gillis1, Ramtin Mehraram1, Robin Lemmens2, Tom Francart1, Maaike Vandermosten1; 1Experimental Oto-Rhino-Laryngology, Department of Neuroscience, Leuven Brain Institute, KU Leuven, Leuven, Belgium, 2Laboratory of Neurobiology, Department of Neuroscience, Leuven Brain Institute, KU Leuven, Leuven, Belgium
People with aphasia (PWA) often have diminished lexical semantic processing as shown by results on behavioral studies as well as event-related potential (ERP) studies, e.g. an altered N400 response. In this study, we will use the neural tracking paradigm, which offers, in contrast to ERPs, a way to study the brain’s response to specific speech representations during continuous speech, hence allowing for an ecologically valid assessment of aphasia. Therefore, we investigated (1) whether neural tracking of two word-level speech representations would differ between a group of 27 PWA (age:72.8±11 y/o) and a group of 22 healthy, age-matched controls (age:71.3±7.4 y/o). Moreover, we explored (2) whether neural tracking can provide information on individual differences in aphasia severity by correlating neural tracking scores with behavioral language test scores. PWA were tested in the chronic phase post-stroke, i.e., min. six months following a left-hemispheric or bilateral lesion. Behavioral testing encompassed a naming test (NBT), a receptive lexico-semantic test (Screeling) and a semantic word fluency test (CAT-NL). EEG was recorded while participants listened to a 24-minute-long Flemish story. The two word-level speech representations of interest were word surprisal and word frequency. The relationship between the EEG signal and these two stimulus-derived speech representations was computed, i.e., the temporal response function (TRF). From the TRF we extracted peak latencies (EEG-outcome-measure-1) and amplitudes (EEG-outcome-measure-2) by searching for local maxima within defined time windows. The TRF was further used for estimating EEG prediction accuracies. To control for the influence of lower-level speech processing, we used the difference in prediction accuracy between two models (see Gillis et al., 2021) (EEG-outcome-measure-3). This outcome measure can be interpreted as unique contribution of word-level speech representations. For all three EEG outcome measures, we performed group comparisons and, within the aphasia group, correlation analyses with the behavioral test scores. These analyses were explorative and thus not corrected for multiple comparisons. The behavioral tests showed significant differences between groups (NBT:W=57.5, p<.001; ScreeLing:W=101, p<.001; CAT-NL:t=-4.94, p<.001). The TRF peak analysis revealed a group difference in the amplitude of the word surprisal peak between 0ms and 265ms in a predefined posterior electrode selection (W=187, p=.027). No other TRF peak showed a group difference in amplitude, nor in latency. Within the aphasia group, the semantic subscale of the receptive lexico-semantic test and the naming test correlated respectively with the peak amplitudes (R=-0.43, p=.04) and latencies (R=0.49, p=.021) of the word frequency peak between 70 and 140ms in a predefined frontal electrode selection. The unique contribution of the word-level representations did reveal neither group differences nor correlations with the behavioral tests. The TRF peak analysis revealed a smaller peak of word surprisal in PWA than in controls. However, word surprisal did not reflect language performance within PWA. Conversely, no group difference was found for word frequency, but it did reflect language performance. A larger sample size may help clarifying these findings. Overall, these results suggest that neural tracking of linguistic speech representations may be a suitable biomarker for the diagnosis of aphasia, but further analyses are needed.
Topic Areas: Disorders: Acquired, Meaning: Lexical Semantics