Comparing language outcomes in monolingual and bilingual stroke patients

Predicting recovery following stroke: Deep learning, multimodal data and feature selection using explainable AI

Machine learning offers great potential for automated prediction of post-stroke symptoms and their response to rehabilitation. Major challenges for this endeavour include the very high dimensionality of neuroimaging data, the relatively small size of the datasets available for learning and interpreting the predictive features, as well as, how to effectively combine neuroimaging and tabular data (e.g. demographic information and clinical characteristics). This paper evaluates several solutions based on two strategies. The first is to use 2D images that summarise MRI scans. The second is to select key features that improve classification accuracy. Additionally, we introduce the novel approach of training a convolutional neural network (CNN) on images that combine regions-of-interests (ROIs) extracted from MRIs, with symbolic representations of tabular data. We evaluate a series of CNN architectures (both 2D and a 3D) that are trained on different representations of MRI and tabular data, to predict whether a composite measure of post-stroke spoken picture description ability is in the aphasic or non-aphasic range. MRI and tabular data were acquired from 758 English speaking stroke survivors who participated in the PLORAS study. Each participant was assigned to one of five different groups that were matched for initial severity of symptoms, recovery time, left lesion size and the months or years post-stroke that spoken description scores were collected. Training and validation were carried out on the first four groups. The fifth (lock-box/test set) group was used to test how well model accuracy generalises to new (unseen) data. The classification accuracy for a baseline logistic regression was 0.678 based on lesion size alone, rising to 0.757 and 0.813 when initial symptom severity and recovery time were successively added. The highest classification accuracy (0.854), area under the curve (0.899) and F1 score (0.901) were observed when 8 regions of interest were extracted from each MRI scan and combined with lesion size, initial severity and recovery time in a 2D Residual Neural Network (ResNet). This was also the best model when data were limited to the 286 participants with moderate or severe initial aphasia (with area under curve = 0.865), a group that would be considered more difficult to classify. Our findings demonstrate how imaging and tabular data can be combined to achieve high post-stroke classification accuracy, even when the dataset is small in machine learning terms. We conclude by proposing how the current models could be improved to achieve even higher levels of accuracy using images from hospital scanners.

Disparity of Primary and Secondary Language Outcomes in Bilingual Patients Undergoing Resection of Glioma of the Speech-Related Regions

BACKGROUND

The existing data about language recovery in bilingual patients come from few studies on acute lesional deficits like stroke or traumatic injury. Still, little is known about the neuroplasticity potential of bilingual patients who undergo resection of gliomas affecting language-eloquent brain regions. In this study, we prospectively evaluated the pre- and postoperative language functions among bilinguals with eloquent region gliomas.

METHODS

We have prospectively collected the preoperative, 3-month and 6-month postoperative data from patients with tumors infiltrating the dominant hemisphere language areas during a 15-month period. Validated Persian/Turkish version of Western Aphasia Battery test and Addenbrooke Cognitive Examination were assessed for main language (L1) and second acquired languages (L2) in each visit.

RESULTS

Twenty-two right-handed bilingual patients were enrolled, and language proficiencies were assessed using mixed model analysis. On baseline and postoperative points, L1 had higher scores in all Addenbrooke Cognitive Examination and Western Aphasia Battery subdomains than L2. Both languages had deterioration at 3-month visit; however, L2 was significantly more deteriorated in all domains. At 6-month visit, both L1 and L2 showed recovery; however, L2 recovered to a less extent than L1. The single most parameter affecting the ultimate language outcome in this study was the preoperative functional level of L1.

CONCLUSIONS

This study shows L1 is less vulnerable to operative insults and L2 may be damaged even when L1 is preserved. We would suggest the more sensitive L2 be used as the screening tool and L1 be used for confirmation of positive responses during language mapping.

Language and Cognitive Outcomes Following Ischemic Stroke in Children With Monolingual and Bilingual Exposure

Aim: Although many children who experience ischemic stroke come from bilingual backgrounds, it is unclear whether bilingual exposure affects poststroke development. Our research evaluates bilingual and monolingual exposure on linguistic/cognitive development poststroke across 3 stroke-onset groups. Method: An institutional stroke registry and medical charts were used to gather data on 237 children across 3 stroke-onset groups: neonatal, <28 days; first-year, 28 days to 12 months; and childhood, 13 months to 18 years. The Pediatric Stroke Outcome Measure (PSOM), administered several times poststroke, was used to evaluate cognition and linguistic development. Results: Similar cognitive outcomes were observed across language groups. However, an interaction effect with stroke-onset group was observed, with monolinguals in the first-year group having worse productive language outcomes as compared to bilinguals. Interpretation: Overall, no detrimental effects of bilingualism were found on children's poststroke cognition and linguistic development. Our study suggests that a bilingual environment may facilitate language development in children poststroke.

Subcortical lesions impact confrontation naming in bilinguals with later age of acquisition: An exploratory study

The bilingual experience is believed to impact brain development and, possibly, cognitive performance. Subcortical structures, including the striatum and white matter, are believed related to confrontation naming performance among bilingual individuals with later age of acquisition (AoA) and lower proficiency of a second language (L2). However, these findings are primarily derived from healthy adult samples, although there is clinical significance for the interpretation of naming performance. The present study examined whether striatal and white matter lesions were associated with naming tasks in clinic-referred bilingual veterans (n = 29) and whether L2 AoA moderated this relationship. Clinically rated lesions, without regard for AoA, were not consistently correlated with naming performance. Moderation models (lesion × AoA) were significant across naming tasks (i.e., naming scores were negatively correlated with striatal lesions with increasing AoA). Effect sizes were higher among striatal models as compared to white matter models. Results extend prior neuroimaging findings with healthy bilinguals that AoA moderates the relationship between subcortical lesions and naming performance in bilingual patients, and suggests that clinicians should consider specifics of bilingual experience when interpreting test scores.

Tree-Based Risk Factor Identification and Stroke Level Prediction in Stroke Cohort Study

Objective. This study focuses on the identification of risk factors, classification of stroke level, and evaluation of the importance and interactions of various patient characteristics using cohort data from the Second Hospital of Lanzhou University. Methodology. Risk factors are identified by evaluation of the relationships between factors and response, as well as by ranking the importance of characteristics. Then, after discarding negligible factors, some well-known multicategorical classification algorithms are used to predict the level of stroke. In addition, using the Shapley additive explanation method (SHAP), factors with positive and negative effects are identified, and some important interactions for classifying the level of stroke are proposed. A waterfall plot for a specific patient is presented and used to determine the risk degree of that patient. Results and Conclusion. The results show that (1) the most important risk factors for stroke are hypertension, history of transient ischemia, and history of stroke; age and gender have a negligible impact. (2) The XGBoost model shows the best performance in predicting stroke risk; it also gives a ranking of risk factors based on their impact. (3) A combination of SHAP and XGBoost can be used to identify positive and negative factors and their interactions in stroke prediction, thereby providing helpful guidance for diagnosis.

Establishing reporting standards for participant characteristics in post-stroke aphasia research: An international e-Delphi exercise and consensus meeting

Objective

To establish international, multidisciplinary expert consensus on minimum participant characteristic reporting standards in aphasia research (DESCRIBE project).

Methods

An international, three-round e-Delphi exercise and consensus meeting, involving multidisciplinary researchers, clinicians and journal editors working academically or clinically in the field of aphasia.

Results

Round 1 of the DESCRIBE e-Delphi exercise ( n = 156) generated 113 items, 20 of which reached consensus by round 3. The final consensus meeting ( n = 19 participants) established DESCRIBE's 14 participant characteristics that should be reported in aphasia studies: age; years of education; biological sex; language of treatment/testing; primary language; languages used; history of condition(s) known to impact communication/cognition; history of previous stroke; lesion hemisphere; time since onset of aphasia; conditions arising from the neurological event; and, for communication partner participants, age, biological sex and relationship to person with aphasia. Each characteristic has been defined and matched with standard response options to enable consistent reporting.

Conclusion

Aphasia research studies should report the 14 DESCRIBE participant characteristics as a minimum. Consistent adherence to the DESCRIBE minimum reporting standard will reduce research wastage and facilitate evidence-based aphasia management by enabling replication and collation of research findings, and translation of evidence into practice.

Efficient and effective assessment of deficits and their neural bases in stroke aphasia

OBJECTIVE

Multi-assessment batteries are necessary for diagnosing and quantifying the multifaceted deficits observed post-stroke. Extensive batteries are thorough but impractically long for clinical settings or large-scale research studies. Clinically-targeted "shallow" batteries superficially cover a wide range of language skills relatively quickly but can struggle to identify mild deficits or quantify the impairment level. Our aim was to compare these batteries across a large group of chronic stroke aphasia and to test a novel data-driven reduced version of an extensive battery that maintained sensitivity to mild impairment, ability to grade deficits and the underlying component structure.

METHODS

We tested 75 chronic left-sided stroke participants, spanning global to mild aphasia. The underlying structure of these three batteries was analysed using cross-validation and principal component analysis, in addition to univariate and multivariate lesion-symptom mapping.

RESULTS

This revealed a four-factor solution for the extensive and data-reduced batteries, identifying phonology, semantic skills, fluency and executive function in contrast to a two-factor solution using the shallow battery (language severity and cognitive severity). Lesion symptom mapping using participants' factor scores identified convergent neural structures for phonology (superior temporal gyrus), semantics (inferior temporal gyrus), speech fluency (precentral gyrus) and executive function (lateral occipitotemporal cortex). The two shallow battery components converged with the phonology and executive function clusters. In addition, we show that multivariate models could predict the component scores using neural data, however not for every component.

CONCLUSIONS

Overall, the data-driven battery appears to be an effective way to save time yet retain maintained sensitivity to mild impairment, ability to grade deficits and the underlying component structure observed in post-stroke aphasia.

Lesion site and therapy time predict responses to a therapy for anomia after stroke: a prognostic model development study

Stroke is a leading cause of disability, and language impairments (aphasia) after stroke are both common and particularly feared. Most stroke survivors with aphasia exhibit anomia (difficulties with naming common objects), but while many therapeutic interventions for anomia have been proposed, treatment effects are typically much larger in some patients than others. Here, we asked whether that variation might be more systematic, and even predictable, than previously thought. 18 patients, each at least 6 months after left hemisphere stroke, engaged in a computerised treatment for their anomia over a 6-week period. Using only: (a) the patients' initial accuracy when naming (to-be) trained items; (b) the hours of therapy that they devoted to the therapy; and (c) whole-brain lesion location data, derived from structural MRI; we developed Partial Least Squares regression models to predict the patients' improvements on treated items, and tested them in cross-validation. Somewhat surprisingly, the best model included only lesion location data and the hours of therapy undertaken. In cross-validation, this model significantly out-performed the null model, in which the prediction for each patient was simply the mean treatment effect of the group. This model also made promisingly accurate predictions in absolute terms: the correlation between empirical and predicted treatment response was 0.62 (95% CI 0.27, 0.95). Our results indicate that individuals' variation in response to anomia treatment are, at least somewhat, systematic and predictable, from the interaction between where and how much lesion damage they have suffered, and the time they devoted to the therapy.

Bilingualism as a protective factor in aphasia

BACKGROUND

Bilingualism may affect the profile of cognitive disturbances associated with stroke. Its impact on aphasia severity, however, is in need of substantiation.

AIMS

To determine the relationship between bilingualism and vascular aphasia severity.

METHODS

This is an observational cross-sectional study conducted on people with post-stroke aphasia. Our sample included 155 monolingual and 53 bilingual vascular aphasia patients. They were selected in a neurological hospital in Kolkata (West Bengal, India).

RESULTS

The Bengali version of Western Aphasia Battery (BWAB) was used in this study. Aphasia severity was compared between monolingual and bilingual participants. The overall difference in the mean aphasia quotient (AQ) between bilingual and monolingual participants was statistically significant (p = 0.009). It was also found that in bilingual participants, aphasia was less severe in certain subgroups, namely: higher lesion volume, male gender, and sub-cortical stroke, while in none of the subgroups a monolingual advantage was documented.

CONCLUSION

Current results suggest that bilingualism represents a protective factor in vascular aphasia; this effect is observed particularly in some aphasia subgroups.

Selective aphasia and focal hypoperfusion in a bilingual patient with HaNDL syndrome

The syndrome of transient headache and neurological deficits with cerebrospinal fluid lymphocytosis (HaNDL) is a rare disease (1) characterized by a benign, self-limited headache syndrome accompanied by neurological deficits (isolated aphasia can be seen in nearly 22% of these patients (2)). Differential diagnosis between acute ischemic stroke and HaNDL syndrome has to be made in order to decide whether to perform a lumbar puncture or start reperfusion treatment early. CT perfusion have proved to be useful for differential diagnosis (3). We present a case of a HaNDL patient referred to the Emergency Department as a stroke in the context of acute onset of selective aphasia (Spanish) in a bilingual patient (French-Spanish). Urgent CT perfusion during the episode revealed increased mean transit time (MTT) with normal Cerebral Blood Flow (CBF) in posterior language areas. The case provides information on a HaNDL attack and its pathophysiology with hemodynamic changes in the acute period during the episode and the benign condition of the illness.

•

Headache and Neurological Deficits with Cerebrospinal Fluid Lymphocytosis (HaNDL) syndrome is a benign and self-limiting clinical condition

•

Differential diagnosis is a challenging scenario in the acute setting.

•

Migraine, stroke, meningoencephalitis or seizures should be considered.

•

Blood flow changes during a HaNDL episode supports common pathophysiological denominator between migraine with aura and HaNDLsyndrome.

•

The case supports the hypothesis that locus for a language learned in later stages is located in an area different from that for the native language.

Investigating the effect of changing parameters when building prediction models for post-stroke aphasia

Neuroimaging has radically improved our understanding of how speech and language abilities map to the brain in normal and impaired participants, including the diverse, graded variations observed in post-stroke aphasia. A handful of studies have begun to explore the reverse inference: creating brain-to-behaviour prediction models. In this study, we explored the effect of three critical parameters on model performance: (1) brain partitions as predictive features, (2) combination of multimodal neuroimaging and (3) type of machine learning algorithms. We explored the influence of these factors while predicting four principal dimensions of language and cognition variation in post-stroke aphasia. Across all four behavioural dimensions, we consistently found that prediction models derived from diffusion-weighted data did not improve performance over models using structural measures extracted from T1 scans. Our results provide a set of principles to guide future work aiming to predict outcomes in neurological patients from brain imaging data.

Generalizing post-stroke prognoses from research data to clinical data

Around a third of stroke survivors suffer from acquired language disorders (aphasia), but current medicine cannot predict whether or when they might recover. Prognostic research in this area increasingly draws on datasets associating structural brain imaging data with outcome scores for ever-larger samples of stroke patients. The aim is to learn brain-behaviour trends from these data, and generalize those trends to predict outcomes for new patients. The practical significance of this work depends on the expected breadth of that generalization. Here, we show that these models can generalize across countries and native languages (from British patients tested in English to Chilean patients tested in Spanish), across neuroimaging technology (from MRI to CT), and from scans collected months or years after stroke for research purposes, to scans collected days or weeks after stroke for clinical purposes. Our results suggest one important confound, in attempting to generalize from research data to clinical data, is the delay between scan acquisition and language assessment. This delay is typically small for research data, where scans and assessments are often acquired contemporaneously. But the most natural, clinical application of these predictions will employ acute prognostic factors to predict much longer-term outcomes. We mitigated this confound by projecting the clinical patients' lesions from the time when their scans were acquired, to the time when their language abilities were assessed; with this projection in place, there was strong evidence that prognoses derived from research data generalized equally well to research and clinical data. These results encourage attention to the confounding role that lesion growth may play in other types of lesion-symptom analysis.

A Comparison of Shallow and Deep Learning Methods for Predicting Cognitive Performance of Stroke Patients From MRI Lesion Images

Stroke causes behavioral deficits in multiple cognitive domains and there is a growing interest in predicting patient performance from neuroimaging data using machine learning techniques. Here, we investigated a deep learning approach based on convolutional neural networks (CNNs) for predicting the severity of language disorder from 3D lesion images from magnetic resonance imaging (MRI) in a heterogeneous sample of stroke patients. CNN performance was compared to that of conventional (shallow) machine learning methods, including ridge regression (RR) on the images' principal components and support vector regression. We also devised a hybrid method based on re-using CNN's high-level features as additional input to the RR model. Predictive accuracy of the four different methods was further investigated in relation to the size of the training set and the level of redundancy across lesion images in the dataset, which was evaluated in terms of location and topological properties of the lesions. The Hybrid model achieved the best performance in most cases, thereby suggesting that the high-level features extracted by CNNs are complementary to principal component analysis features and improve the model's predictive accuracy. Moreover, our analyses indicate that both the size of training data and image redundancy are critical factors in determining the accuracy of a computational model in predicting behavioral outcome from the structural brain imaging data of stroke patients.

Significance of speech production errors on cross-linguistic processing in Sepedi-English individuals with bilingual aphasia: a case series analysis

BACKGROUND

Bilingual aphasia forms a significant part of speech-language pathologists' (SLP) caseload, globally, and specifically in South Africa. Few tools supporting clinical decision-making are available due to limited understanding of typical and disordered cross-linguistic processing (how the languages interact). Speech errors may provide insight about "hidden" bilingual mechanisms.

OBJECTIVES

To determine what speech errors can impart about cross-linguistic processing, as well as, associated language and impairment variables in Sepedi-English individuals with aphasia.

METHOD

The case series included six participants, purposively selected from three rehabilitation sites in South Africa. Detailed language and clinical profiles were obtained. Participants performed a confrontation naming task in their most dominant (MDL) and less dominant language (LDL). Responses were audio recorded, transcribed, and analyzed for overall accuracy and error type in MDL and LDL; verified by a Sepedi-speaking linguist and a qualified SLP.

RESULTS

(1) No statistically significant differences in MDL-LDL naming accuracy were found, supporting recent literature of simultaneous inter-activation of both languages and shared word retrieval mechanisms. All types of speech errors occurred, and semantic errors were produced most frequently and consistently in each participant's MDL and LDL. (2) Language proficiency, language recovery patterns, and aphasia type (Broca's and Anomic) and severity (mild and/or moderate) appeared to be more strongly linked to cross-linguistic processing than Sepedi-English linguistic differences and age of acquisition of both languages.

CONCLUSIONS

Participants with bilingual aphasia may use typical cross-linguistic and word retrieval mechanisms, concurring with current theories of bilingualism. Findings are preliminary, warranting investigations of other language tasks, modalities, pairs, and related variables.

fMRI data processing in MRTOOL: to what extent does anatomical registration affect the reliability of functional results?

Spatial registration is an essential step in the analysis of fMRI data because it enables between-subject analyses of brain activity, measured either during task performance or in the resting state. In this study, we investigated how anatomical registration with MRTOOL affects the reliability of task-related fMRI activity. We used as a benchmark the results from two other spatial registration methods implemented in SPM12: the Unified Segmentation algorithm and the DARTEL toolbox. Structural alignment accuracy and the impact on functional activation maps were assessed with high-resolution T1-weighted images and a set of task-related functional volumes acquired in 10 healthy volunteers. Our findings confirmed that anatomical registration is a crucial step in fMRI data processing, contributing significantly to the total inter-subject variance of the activation maps. MRTOOL and DARTEL provided greater registration accuracy than Unified Segmentation. Although DARTEL had superior gray matter and white matter tissue alignment than MRTOOL, there were no significant differences between DARTEL and MRTOOL in test-retest reliability. Likewise, we found only limited differences in BOLD activation morphology between MRTOOL and DARTEL. The test-retest reliability of task-related responses was comparable between MRTOOL and DARTEL, and both proved superior to Unified Segmentation. We conclude that MRTOOL, which is suitable for single-subject processing of structural and functional MR images, is a valid alternative to other SPM12-based approaches that are intended for group analysis. MRTOOL now includes a normalization module for fMRI data and is freely available to the scientific community.

The impact of sample size on the reproducibility of voxel-based lesion-deficit mappings

This study investigated how sample size affects the reproducibility of findings from univariate voxel-based lesion-deficit analyses (e.g., voxel-based lesion-symptom mapping and voxel-based morphometry). Our effect of interest was the strength of the mapping between brain damage and speech articulation difficulties, as measured in terms of the proportion of variance explained. First, we identified a region of interest by searching on a voxel-by-voxel basis for brain areas where greater lesion load was associated with poorer speech articulation using a large sample of 360 right-handed English-speaking stroke survivors. We then randomly drew thousands of bootstrap samples from this data set that included either 30, 60, 90, 120, 180, or 360 patients. For each resample, we recorded effect size estimates and p values after conducting exactly the same lesion-deficit analysis within the previously identified region of interest and holding all procedures constant. The results show (1) how often small effect sizes in a heterogeneous population fail to be detected; (2) how effect size and its statistical significance varies with sample size; (3) how low-powered studies (due to small sample sizes) can greatly over-estimate as well as under-estimate effect sizes; and (4) how large sample sizes (N ≥ 90) can yield highly significant p values even when effect sizes are so small that they become trivial in practical terms. The implications of these findings for interpreting the results from univariate voxel-based lesion-deficit analyses are discussed.

Predicting language outcomes after stroke: Is structural disconnection a useful predictor?

For many years, researchers have sought to understand whether and when stroke survivors with acquired language impairment (aphasia) will recover. There is broad agreement that lesion location information should play some role in these predictions, but still no consensus on the best or right way to encode that information. Here, we address the emerging emphasis on the structural connectome in this work - specifically the claim that disrupted white matter connectivity conveys important, unique prognostic information for stroke survivors with aphasia. Our sample included 818 stroke patients extracted from the PLORAS database, which associates structural MRI from stroke patients with language assessment scores from the Comprehensive Aphasia Test (CAT) and basic demographic. Patients were excluded when their lesions were too diffuse or small (<1 cm3) to be detected by the Automatic Lesion Identification toolbox, which we used to encode patients' lesions as binary lesion images in standard space. Lesions were encoded using the 116 regions defined by the Automatic Anatomical Labelling atlas. We examined prognostic models driven by both "lesion load" in these regions (i.e. the proportion of each region destroyed by each patient's lesion), and by the disconnection of the white matter connections between them which was calculated via the Network Modification toolbox. Using these data, we build a series of prognostic models to predict first one ("naming"), and then all of the language scores defined by the CAT. We found no consistent evidence that connectivity disruption data in these models improved our ability to predict any language score. This may be because the connectivity disruption variables are strongly correlated with the lesion load variables: correlations which we measure both between pairs of variables in their original form, and between principal components of both datasets. Our conclusion is that, while both types of structural brain data do convey useful, prognostic information in this domain, they also appear to convey largely the same variance. We conclude that connectivity disruption variables do not help us to predict patients' language skills more accurately than lesion location (load) data alone.

Predicting the pattern and severity of chronic post-stroke language deficits from functionally-partitioned structural lesions

There is an ever-increasing wealth of knowledge arising from basic cognitive and clinical neuroscience on how speech and language capabilities are organised in the brain. It is, therefore, timely to use this accumulated knowledge and expertise to address critical research challenges, including the ability to predict the pattern and level of language deficits found in aphasic patients (a third of all stroke cases). Previous studies have mainly focused on discriminating between broad aphasia dichotomies from purely anatomically-defined lesion information. In the current study, we developed and assessed a novel approach in which core language areas were mapped using principal component analysis in combination with correlational lesion mapping and the resultant ‘functionally-partitioned’ lesion maps were used to predict a battery of 21 individual test scores as well as aphasia subtype for 70 patients with chronic post-stroke aphasia. Specifically, we used lesion information to predict behavioural scores in regression models (cross-validated using 5-folds). The winning model was identified through the adjusted R² (model fit to data) and performance in predicting holdout folds (generalisation to new cases). We also used logistic regression to predict fluent/non-fluent status and aphasia subtype. Functionally-partitioned models generally outperformed other models at predicting individual tests, fluency status and aphasia subtype.

•

Predict the pattern and level of language deficits found in chronic aphasic patients

•

Use principal component analysis to identify functional lesion maps

•

Functionally-partitioned lesion maps used as predictor variables instead of lesion volume

•

Functionally-partitioned lesion model plus age produced the best regression model

•

Model can successfully predict fluent/non-fluent types and aphasia classification

How distributed processing produces false negatives in voxel-based lesion-deficit analyses

In this study, we hypothesized that if the same deficit can be caused by damage to one or another part of a distributed neural system, then voxel-based analyses might miss critical lesion sites because preservation of each site will not be consistently associated with preserved function. The first part of our investigation used voxel-based multiple regression analyses of data from 359 right-handed stroke survivors to identify brain regions where lesion load is associated with picture naming abilities after factoring out variance related to object recognition, semantics and speech articulation so as to focus on deficits arising at the word retrieval level. A highly significant lesion-deficit relationship was identified in left temporal and frontal/premotor regions. Post-hoc analyses showed that damage to either of these sites caused the deficit of interest in less than half the affected patients (76/162 = 47%). After excluding all patients with damage to one or both of the identified regions, our second analysis revealed a new region, in the anterior part of the left putamen, which had not been previously detected because many patients had the deficit of interest after temporal or frontal damage that preserved the left putamen. The results illustrate how (i) false negative results arise when the same deficit can be caused by different lesion sites; (ii) some of the missed effects can be unveiled by adopting an iterative approach that systematically excludes patients with lesions to the areas identified in previous analyses, (iii) statistically significant voxel-based lesion-deficit mappings can be driven by a subset of patients; (iv) focal lesions to the identified regions are needed to determine whether the deficit of interest is the consequence of focal damage or much more extensive damage that includes the identified region; and, finally, (v) univariate voxel-based lesion-deficit mappings cannot, in isolation, be used to predict outcome in other patients.

Using multi-level Bayesian lesion-symptom mapping to probe the body-part-specificity of gesture imitation skills

Past attempts to identify the neural substrates of hand and finger imitation skills in the left hemisphere of the brain have yielded inconsistent results. Here, we analyse those associations in a large sample of 257 left hemisphere stroke patients. By introducing novel Bayesian methods, we characterise lesion symptom associations at three levels: the voxel-level, the single-region level (using anatomically defined regions), and the region-pair level. The results are inconsistent across those three levels and we argue that each level of analysis makes assumptions which constrain the results it can produce. Regardless of the inconsistencies across levels, and contrary to past studies which implicated differential neural substrates for hand and finger imitation, we find no consistent voxels or regions, where damage affects one imitation skill and not the other, at any of the three analysis levels. Our novel Bayesian approach indicates that any apparent differences appear to be driven by an increased sensitivity of hand imitation skills to lesions that also impair finger imitation. In our analyses, the results of the highest level of analysis (region-pairs) emphasise a role of the primary somatosensory and motor cortices, and the occipital lobe in imitation. We argue that this emphasis supports an account of both imitation tasks based on direct sensor-motor connections, which throws doubt on past accounts which imply the need for an intermediate (e.g. body-part-coding) system of representation.

Right hemisphere structural adaptation and changing language skills years after left hemisphere stroke

Stroke survivors with acquired language deficits are commonly thought to reach a ‘plateau’ within a year of stroke onset, after which their residual language skills will remain stable. Nevertheless, there have been reports of patients who appear to recover over years. Here, we analysed longitudinal change in 28 left-hemisphere stroke patients, each more than a year post-stroke when first assessed—testing each patient’s spoken object naming skills and acquiring structural brain scans twice. Some of the patients appeared to improve over time while others declined; both directions of change were associated with, and predictable given, structural adaptation in the intact right hemisphere of the brain. Contrary to the prevailing view that these patients’ language skills are stable, these results imply that real change continues over years. The strongest brain–behaviour associations (the ‘peak clusters’) were in the anterior temporal lobe and the precentral gyrus. Using functional magnetic resonance imaging, we confirmed that both regions are actively involved when neurologically normal control subjects name visually presented objects, but neither appeared to be involved when the same participants used a finger press to make semantic association decisions on the same stimuli. This suggests that these regions serve word-retrieval or articulatory functions in the undamaged brain. We teased these interpretations apart by reference to change in other tasks. Consistent with the claim that the real change is occurring here, change in spoken object naming was correlated with change in two other similar tasks, spoken action naming and written object naming, each of which was independently associated with structural adaptation in similar (overlapping) right hemisphere regions. Change in written object naming, which requires word-retrieval but not articulation, was also significantly more correlated with both (i) change in spoken object naming; and (ii) structural adaptation in the two peak clusters, than was change in another task—auditory word repetition—which requires articulation but not word retrieval. This suggests that the changes in spoken object naming reflected variation at the level of word-retrieval processes. Surprisingly, given their qualitatively similar activation profiles, hypertrophy in the anterior temporal region was associated with improving behaviour, while hypertrophy in the precentral gyrus was associated with declining behaviour. We predict that either or both of these regions might be fruitful targets for neural stimulation studies (suppressing the precentral region and/or enhancing the anterior temporal region), aiming to encourage recovery or arrest decline even years after stroke occurs.

Distinct distributed patterns of neural activity are associated with two languages in the bilingual brain

A large body of previous neuroimaging studies suggests that multiple languages are processed and organized in a single neuroanatomical system in the bilingual brain, although differential activation may be seen in some studies because of different proficiency levels and/or age of acquisition of the two languages. However, one important possibility is that the two languages may involve interleaved but functionally independent neural populations within a given cortical region, and thus, distinct patterns of neural computations may be pivotal for the processing of the two languages. Using functional magnetic resonance imaging (fMRI) and multivariate pattern analyses, we tested this possibility in Chinese-English bilinguals when they performed an implicit reading task. We found a broad network of regions wherein the two languages evoked different patterns of activity, with only partially overlapping patterns of voxels in a given region. These regions, including the middle occipital cortices, fusiform gyri, and lateral temporal, temporoparietal, and prefrontal cortices, are associated with multiple aspects of language processing. The results suggest the functional independence of neural computations underlying the representations of different languages in bilinguals.

Ten problems and solutions when predicting individual outcome from lesion site after stroke

In this paper, we consider solutions to ten of the challenges faced when trying to predict an individual's functional outcome after stroke on the basis of lesion site. A primary goal is to find lesion-outcome associations that are consistently observed in large populations of stroke patients because consistent associations maximise confidence in future individualised predictions. To understand and control multiple sources of inter-patient variability, we need to systematically investigate each contributing factor and how each factor depends on other factors. This requires very large cohorts of patients, who differ from one another in typical and measurable ways, including lesion site, lesion size, functional outcome and time post stroke (weeks to decades). These multivariate investigations are complex, particularly when the contributions of different variables interact with one another. Machine learning algorithms can help to identify the most influential variables and indicate dependencies between different factors. Multivariate lesion analyses are needed to understand how the effect of damage to one brain region depends on damage or preservation in other brain regions. Such data-led investigations can reveal predictive relationships between lesion site and outcome. However, to understand and improve the predictions we need explanatory models of the neural networks and degenerate pathways that support functions of interest. This will entail integrating the results of lesion analyses with those from functional imaging (fMRI, MEG), transcranial magnetic stimulation (TMS) and diffusor tensor imaging (DTI) studies of healthy participants and patients.

Decoding post-stroke motor function from structural brain imaging

Clinical research based on neuroimaging data has benefited from machine learning methods, which have the ability to provide individualized predictions and to account for the interaction among units of information in the brain. Application of machine learning in structural imaging to investigate diseases that involve brain injury presents an additional challenge, especially in conditions like stroke, due to the high variability across patients regarding characteristics of the lesions. Extracting data from anatomical images in a way that translates brain damage information into features to be used as input to learning algorithms is still an open question. One of the most common approaches to capture regional information from brain injury is to obtain the lesion load per region (i.e. the proportion of voxels in anatomical structures that are considered to be damaged). However, no systematic evaluation has yet been performed to compare this approach with using patterns of voxels (i.e. considering each voxel as a single feature). In this paper we compared both approaches applying Gaussian Process Regression to decode motor scores in 50 chronic stroke patients based solely on data derived from structural MRI. For both approaches we compared different ways to delimit anatomical areas: regions of interest from an anatomical atlas, the corticospinal tract, a mask obtained from fMRI analysis with a motor task in healthy controls and regions selected using lesion-symptom mapping. Our analysis showed that extracting features through patterns of voxels that represent lesion probability produced better results than quantifying the lesion load per region. In particular, from the different ways to delimit anatomical areas compared, the best performance was obtained with a combination of a range of cortical and subcortical motor areas as well as the corticospinal tract. These results will inform the appropriate methodology for predicting long term motor outcomes from early post-stroke structural brain imaging.

The PLORAS Database: A data repository for Predicting Language Outcome and Recovery After Stroke

The PLORAS Database is a relational repository of anatomical and functional imaging data that has primarily been acquired from stroke survivors, along with standardized scores on a wide range of sensory, motor and cognitive abilities, demographic details and medical history. As of January 2015, we have data from 750 patients with an expected accrual rate of 200 patients per year. Expansion will accelerate as we extend our collaborations. The main aim of the database is to Predict Language Outcome and Recovery After Stroke (PLORAS) on the basis of a single structural (anatomical) brain scan that indexes the stereotactic location and extent of brain damage. Predictions are made for individual patients by indicating how other patients with the most similar brain damage, cognitive abilities and demographic details recovered their language skills over time. Predictions are validated by longitudinal follow-ups of patients who initially presented with speech and language difficulties. The PLORAS Database can also be used to predict recovery of other cognitive abilities on the basis of anatomical brain scans. The functional imaging data can be used to understand the neural mechanisms that support recovery from brain damage; and all the data can be used to understand the main sources of inter-subject variability in structure–function mappings in the human brain. Data will be made available for sharing, subject to: funding, ethical approval and patient consent.

•

The PLORAS Database is a repository of data from hundreds of stroke patients.

•

Lesion site is identified from T1-weighted structural MRI scans.

•

Impairments are assessed using the Comprehensive Aphasia Test.

•

Functional MRI data are collected from 14 different speech and language tasks.

•

All data contribute to understanding and modeling inter-subject variability.