Machine learning-based multimodal prediction of language outcomes in chronic aphasia

Dissociation of White Matter Bundles in Different Recovery Measures in Poststroke Aphasia

BACKGROUND

Poststroke aphasia (PSA) recovery shows high variability across individuals and at different time points. Although diffusion biomarkers from the ventral and dorsal streams have demonstrated strong predictive power for language outcomes, it is still unclear how these biomarkers relate to the various stages of PSA recovery. In this study, we aim to compare diffusion metrics and language measures as predictors of language recovery in a longitudinal cohort of participants with PSA.

METHODS

Participants were recruited at a stroke unit at the emergency room, and underwent diffusion magnetic resonance imaging scanning and language assessment within 3 days (acute phase) after stroke, with behavioral follow-ups at subacute (10±3 days) and chronic phases (>6 months). We conducted regression analyses on language performance (cross-sectional), Δscores between all time points (acute-subacute, subacute-chronic, acute-chronic), and relative Δscores between all time points (Δscore/language baseline score), with acute diffusion metrics from language-related white matter tracts, lesion size, language baseline scores, and demographic data as predictors.

RESULTS

Thirty-nine participants presenting PSA were recruited, and 24 participants (mean age, 73 years; 8 women) completed the 3-time point assessment in total. The best prediction model of performance scores used axial diffusivity from the left arcuate fasciculus in both the subacute (R2=0.785) and chronic stages (R2=0.626). Moreover, the prediction of ∆scores depended on axial diffusivity from the left inferior frontal-occipital fasciculus in the subacute stage (R2=0.5) and depended additionally on axial diffusivity from the right inferior frontal-occipital fasciculus in the chronic stage (R2=0.68). The prediction of mediation analyses showed that the lesion load of the left arcuate fasciculus mediated the relationship between axial diffusivity from the left arcuate fasciculus and chronic language performance.

CONCLUSIONS

Language performance at subacute and chronic time points could be predicted by the integrity of the left arcuate fasciculus, whereas Δscores in the subacute and chronic phases depended on the left inferior frontal-occipital fasciculus, showing a dissociation of the white matter pathways about language outcomes. These results suggest a functional differentiation of the dual-stream components in PSA recovery.

A Lesion-aware Edge-based Graph Neural Network for Predicting Language Ability in Patients with Post-stroke Aphasia

We propose a lesion-aware graph neural network (LEGNet) to predict language ability from resting-state fMRI (rs-fMRI) connectivity in patients with post-stroke aphasia. Our model integrates three components: an edge-based learning module that encodes functional connectivity between brain regions, a lesion encoding module, and a subgraph learning module that leverages functional similarities for prediction. We use synthetic data derived from the Human Connectome Project (HCP) for hyperparameter tuning and model pretraining. We then evaluate the performance using repeated 10-fold cross-validation on an in-house neuroimaging dataset of post-stroke aphasia. Our results demonstrate that LEGNet outperforms baseline deep learning methods in predicting language ability. LEGNet also exhibits superior generalization ability when tested on a second in-house dataset that was acquired under a slightly different neuroimaging protocol. Taken together, the results of this study highlight the potential of LEGNet in effectively learning the relationships between rs-fMRI connectivity and language ability in a patient cohort with brain lesions for improved post-stroke aphasia evaluation.

AI-assisted assessment and treatment of aphasia: a review

Aphasia is a language disorder caused by brain injury that often results in difficulties with speech production and comprehension, significantly impacting the affected individuals' lives. Recently, artificial intelligence (AI) has been advancing in medical research. Utilizing machine learning and related technologies, AI develops sophisticated algorithms and predictive models, and can employ tools such as speech recognition and natural language processing to autonomously identify and analyze language deficits in individuals with aphasia. These advancements provide new insights and methods for assessing and treating aphasia. This article explores current AI-supported assessment and treatment approaches for aphasia and highlights key application areas. It aims to uncover how AI can enhance the process of assessment, tailor therapeutic interventions, and track the progress and outcomes of rehabilitation efforts. The article also addresses the current limitations of AI's application in aphasia and discusses prospects for future research.

Assessing Relative Linguistic Impairment With Model-Based Item Selection

PURPOSE

A picture naming test is presented that reveals impairment to specific mechanisms involved in the naming process, using accuracy scores on curated item sets. A series of psychometric validation experiments are reported.

METHOD

Using a computational model that enables estimation of item difficulty at the lexical and sublexical stages of word retrieval, two complimentary sets of items were constructed that challenge the respective psycholinguistic levels of representation. The difference in accuracy between these item sets yields the relative linguistic impairment (RLI) score. In a cohort of 91 people with chronic left-hemisphere stroke who enrolled in a clinical trial for anomia, we assessed psychometric properties of the RLI score and then used the new scale to make predictions about other language behaviors, lesion distributions, and functional activation during naming.

RESULTS

RLI scores had adequate psychometric properties for clinical significance. RLI scores contained predictive information about spontaneous speech fluency, over and above accuracy. A dissociation was observed between performance on the RLI item sets and performance on the subtests of an independent language battery. Sublexical RLI was significantly associated with apraxia of speech and with lesions encompassing perisylvian regions, while lexical RLI was associated with lesions to deep white matter. The RLI construct was reflected in functional brain activity during naming, independent of overall accuracy, with a respective shift of activation between dorsal and ventral networks responsible for different aspects of word retrieval.

CONCLUSION

The RLI assessment satisfies the psychometric requirements to serve as a useful clinical measure.

Connected Speech Fluency in Poststroke and Progressive Aphasia: A Scoping Review of Quantitative Approaches and Features

PURPOSE

Speech fluency has important diagnostic implications for individuals with poststroke aphasia (PSA) as well as primary progressive aphasia (PPA), and quantitative assessment of connected speech has emerged as a widely used approach across both etiologies. The purpose of this review was to provide a clearer picture on the range, nature, and utility of individual quantitative speech/language measures and methods used to assess connected speech fluency in PSA and PPA, and to compare approaches across etiologies.

METHOD

We conducted a scoping review of literature published between 2012 and 2022 following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews guidelines. Forty-five studies were included in the review. Literature was charted and summarized by etiology and characteristics of included patient populations and method(s) used for derivation and analysis of speech/language features. For a subset of included articles, we also charted the individual quantitative speech/language features reported and the level of significance of reported results.

RESULTS

Results showed that similar methodological approaches have been used to quantify connected speech fluency in both PSA and PPA. Two hundred nine individual speech-language features were analyzed in total, with low levels of convergence across etiology on specific features but greater agreement on the most salient features. The most useful features for differentiating fluent from nonfluent aphasia in both PSA and PPA were features related to overall speech quantity, speech rate, or grammatical competence.

CONCLUSIONS

Data from this review demonstrate the feasibility and utility of quantitative approaches to index connected speech fluency in PSA and PPA. We identified emergent trends toward automated analysis methods and data-driven approaches, which offer promising avenues for clinical translation of quantitative approaches. There is a further need for improved consensus on which subset of individual features might be most clinically useful for assessment and monitoring of fluency.

SUPPLEMENTAL MATERIAL

https://doi.org/10.23641/asha.25537237.

Distinct brain morphometry patterns revealed by deep learning improve prediction of post-stroke aphasia severity

BACKGROUND

Emerging evidence suggests that post-stroke aphasia severity depends on the integrity of the brain beyond the lesion. While measures of lesion anatomy and brain integrity combine synergistically to explain aphasic symptoms, substantial interindividual variability remains unaccounted. One explanatory factor may be the spatial distribution of morphometry beyond the lesion (e.g., atrophy), including not just specific brain areas, but distinct three-dimensional patterns.

METHODS

Here, we test whether deep learning with Convolutional Neural Networks (CNNs) on whole brain morphometry (i.e., segmented tissue volumes) and lesion anatomy better predicts chronic stroke individuals with severe aphasia (N = 231) than classical machine learning (Support Vector Machines; SVMs), evaluating whether encoding spatial dependencies identifies uniquely predictive patterns.

RESULTS

CNNs achieve higher balanced accuracy and F1 scores, even when SVMs are nonlinear or integrate linear or nonlinear dimensionality reduction. Parity only occurs when SVMs access features learned by CNNs. Saliency maps demonstrate that CNNs leverage distributed morphometry patterns, whereas SVMs focus on the area around the lesion. Ensemble clustering of CNN saliencies reveals distinct morphometry patterns unrelated to lesion size, consistent across individuals, and which implicate unique networks associated with different cognitive processes as measured by the wider neuroimaging literature. Individualized predictions depend on both ipsilateral and contralateral features outside the lesion.

CONCLUSIONS

Three-dimensional network distributions of morphometry are directly associated with aphasia severity, underscoring the potential for CNNs to improve outcome prognostication from neuroimaging data, and highlighting the prospective benefits of interrogating spatial dependence at different scales in multivariate feature space.

Large language models in physical therapy: time to adapt and adept

Healthcare is experiencing a transformative phase, with artificial intelligence (AI) and machine learning (ML). Physical therapists (PTs) stand on the brink of a paradigm shift in education, practice, and research. Rather than visualizing AI as a threat, it presents an opportunity to revolutionize. This paper examines how large language models (LLMs), such as ChatGPT and BioMedLM, driven by deep ML can offer human-like performance but face challenges in accuracy due to vast data in PT and rehabilitation practice. PTs can benefit by developing and training an LLM specifically for streamlining administrative tasks, connecting globally, and customizing treatments using LLMs. However, human touch and creativity remain invaluable. This paper urges PTs to engage in learning and shaping AI models by highlighting the need for ethical use and human supervision to address potential biases. Embracing AI as a contributor, and not just a user, is crucial by integrating AI, fostering collaboration for a future in which AI enriches the PT field provided data accuracy, and the challenges associated with feeding the AI model are sensitively addressed.

AI and Aphasia in the Digital Age: A Critical Review

Aphasiology has a long and rich tradition of contributing to understanding how culture, language, and social environment contribute to brain development and function. Recent breakthroughs in AI can transform the role of aphasiology in the digital age by leveraging speech data in all languages to model how damage to specific brain regions impacts linguistic universals such as grammar. These tools, including generative AI (ChatGPT) and natural language processing (NLP) models, could also inform practitioners working with clinical populations in the assessment and treatment of aphasia using AI-based interventions such as personalized therapy and adaptive platforms. Although these possibilities have generated enthusiasm in aphasiology, a rigorous interrogation of their limitations is necessary before AI is integrated into practice. We explain the history and first principles of reciprocity between AI and aphasiology, highlighting how lesioning neural networks opened the black box of cognitive neurolinguistic processing. We then argue that when more data from aphasia across languages become digitized and available online, deep learning will reveal hitherto unreported patterns of language processing of theoretical interest for aphasiologists. We also anticipate some problems using AI, including language biases, cultural, ethical, and scientific limitations, a misrepresentation of marginalized languages, and a lack of rigorous validation of tools. However, as these challenges are met with better governance, AI could have an equitable impact.

Discourse- and lesion-based aphasia quotient estimation using machine learning

Discourse is a fundamentally important aspect of communication, and discourse production provides a wealth of information about linguistic ability. Aphasia commonly affects, in multiple ways, the ability to produce discourse. Comprehensive aphasia assessments such as the Western Aphasia Battery-Revised (WAB-R) are time- and resource-intensive. We examined whether discourse measures can be used to estimate WAB-R Aphasia Quotient (AQ), and whether this can serve as an ecologically valid, less resource-intensive measure. We used features extracted from discourse tasks using three AphasiaBank prompts involving expositional (picture description), story narrative, and procedural discourse. These features were used to train a machine learning model to predict the WAB-R AQ. We also compared and supplemented the model with lesion location information from structural neuroimaging. We found that discourse-based models could estimate AQ well, and that they outperformed models based on lesion features. Addition of lesion features to the discourse features did not improve the performance of the discourse model substantially. Inspection of the most informative discourse features revealed that different prompt types taxed different aspects of language. These findings suggest that discourse can be used to estimate aphasia severity, and provide insight into the linguistic content elicited by different types of discourse prompts.

From concept to practice: a scoping review of the application of AI to aphasia diagnosis and management

PURPOSE

Aphasia is an acquired communication disability resulting from impairments in language processing following brain injury, most commonly stroke. People with aphasia experience difficulties in all modalities of language that impact their quality of life. Therefore, researchers have investigated the use of Artificial Intelligence (AI) to deliver innovative solutions in Aphasia management and rehabilitation.

MATERIALS AND METHODS

We conducted a scoping review of the use of AI in aphasia research and rehabilitation to explore the evolution of AI applications to aphasia, the progression of technologies and applications. Furthermore, we aimed to identify gaps in the use of AI in Aphasia to highlight the potential areas where AI might add value. We analysed 77 studies to determine the research objectives, the history of AI techniques in Aphasia and their progression over time.

RESULTS

Most of the studies focus on automated assessment using AI, with recent studies focusing on AI for therapy and personalised assistive systems. Starting from prototypes and simulations, the use of AI has progressed to include supervised machine learning, unsupervised machine learning, natural language processing, fuzzy rules, and genetic programming.

CONCLUSION

Considerable scope remains to align AI technology with aphasia rehabilitation to empower patient-centred, customised rehabilitation and enhanced self-management.

Multivariate lesion symptom mapping for predicting trajectories of recovery from aphasia

Individuals with post-stroke aphasia tend to recover their language to some extent; however, it remains challenging to reliably predict the nature and extent of recovery that will occur in the long term. The aim of this study was to quantitatively predict language outcomes in the first year of recovery from aphasia across multiple domains of language and at multiple timepoints post-stroke. We recruited 217 patients with aphasia following acute left hemisphere ischaemic or haemorrhagic stroke and evaluated their speech and language function using the Quick Aphasia Battery acutely and then acquired longitudinal follow-up data at up to three timepoints post-stroke: 1 month (n = 102), 3 months (n = 98) and 1 year (n = 74). We used support vector regression to predict language outcomes at each timepoint using acute clinical imaging data, demographic variables and initial aphasia severity as input. We found that ∼60% of the variance in long-term (1 year) aphasia severity could be predicted using these models, with detailed information about lesion location importantly contributing to these predictions. Predictions at the 1- and 3-month timepoints were somewhat less accurate based on lesion location alone, but reached comparable accuracy to predictions at the 1-year timepoint when initial aphasia severity was included in the models. Specific subdomains of language besides overall severity were predicted with varying but often similar degrees of accuracy. Our findings demonstrate the feasibility of using support vector regression models with leave-one-out cross-validation to make personalized predictions about long-term recovery from aphasia and provide a valuable neuroanatomical baseline upon which to build future models incorporating information beyond neuroanatomical and demographic predictors.

Insights gained over 60 years on factors shaping post-stroke aphasia recovery: A commentary on Vignolo (1964)

Aphasia is an acquired language disorder resulting from brain injury, including strokes which is the most common etiology, neurodegenerative diseases, tumors, traumatic brain injury, and resective surgery. Aphasia affects a significant portion of stroke survivors, with approximately one third experiencing its debilitating effects in the long term. Despite its challenges, there is growing evidence that recovery from aphasia is possible, even in the chronic phase of stroke. Sixty years ago, Vignolo (1964) outlined the primary challenges confronted by researchers in this field. These challenges encompassed the absence of an objective evaluation of language difficulties, the scarcity of evidence regarding spontaneous aphasia recovery, and the presence of numerous variables that could potentially influence the process of aphasia recovery. In this paper, we discuss the remarkable progress that has been made in the assessment of language and communication in aphasia as well as in understanding the factors influencing post-stroke aphasia recovery.

Cerebral perfusion in post-stroke aphasia and its relationship to residual language abilities

Stroke alters blood flow to the brain resulting in damaged tissue and cell death. Moreover, the disruption of cerebral blood flow (perfusion) can be observed in areas surrounding and distal to the lesion. These structurally preserved but suboptimally perfused regions may also affect recovery. Thus, to better understand aphasia recovery, the relationship between cerebral perfusion and language needs to be systematically examined. In the current study, we aimed to evaluate (i) how stroke affects perfusion outside of lesioned areas in chronic aphasia and (ii) how perfusion in specific cortical areas and perilesional tissue relates to language outcomes in aphasia. We analysed perfusion data from a large sample of participants with chronic aphasia due to left hemisphere stroke (n = 43) and age-matched healthy controls (n = 25). We used anatomically defined regions of interest that covered the frontal, parietal, and temporal areas of the perisylvian cortex in both hemispheres, areas typically known to support language, along with several control regions not implicated in language processing. For the aphasia group, we also looked at three regions of interest in the perilesional tissue. We compared perfusion levels between the two groups and investigated the relationship between perfusion levels and language subtest scores while controlling for demographic and lesion variables. First, we observed that perfusion levels outside the lesioned areas were significantly reduced in frontal and parietal regions in the left hemisphere in people with aphasia compared to the control group, while no differences were observed for the right hemisphere regions. Second, we found that perfusion in the left temporal lobe (and most strongly in the posterior part of both superior and middle temporal gyri) and inferior parietal areas (supramarginal gyrus) was significantly related to residual expressive and receptive language abilities. In contrast, perfusion in the frontal regions did not show such a relationship; no relationship with language was also observed for perfusion levels in control areas and all right hemisphere regions. Third, perilesional perfusion was only marginally related to language production abilities. Cumulatively, the current findings demonstrate that blood flow is reduced beyond the lesion site in chronic aphasia and that hypoperfused neural tissue in critical temporoparietal language areas has a negative impact on behavioural outcomes. These results, using perfusion imaging, underscore the critical and general role that left hemisphere posterior temporal regions play in various expressive and receptive language abilities. Overall, the study highlights the importance of exploring perfusion measures in stroke.

Cerebral blood flow in patients recovered from mild COVID-19

BACKGROUND AND PURPOSE

Cerebral hypoperfusion has been described in both severe and mild forms of symptomatic Coronavirus Disease 2019 (COVID-19) infection. The purpose of this study was to investigate global and regional cerebral blood flow (CBF) in asymptomatic COVID-19 patients.

METHODS

Cases with mild COVID-19 infection and age-, sex-, and race-matched healthy controls were drawn from the Aging Brain Consortium at The University of South Carolina data repository. Demographics, risk factors, and data from the Montreal Cognitive Assessment were collected. Mean CBF values for gray matter (GM), white matter (WM), and the whole brain were calculated by averaging CBF values of standard space-normalized CBF image values falling within GM and WM masks. Whole brain region of interest-based analyses were used to create standardized CBF maps and explore differences between groups.

RESULTS

Twenty-eight cases with prior mild COVID-19 infection were compared with 28 controls. Whole-brain CBF (46.7 ± 5.6 vs. 49.3 ± 3.7, p = .05) and WM CBF (29.3 ± 2.6 vs. 31.0 ± 1.6, p = .03) were noted to be significantly lower in COVID-19 cases as compared to controls. Predictive models based on these data predicted COVID-19 group membership with a high degree of accuracy (85.2%, p < .001), suggesting CBF patterns are an imaging marker of mild COVID-19 infection.

CONCLUSION

In this study, lower WM CBF, as well as widespread regional CBF changes identified using quantitative MRI, was found in mild COVID-19 patients. Further studies are needed to determine the reliability of this newly identified COVID-19 brain imaging marker and determine what drives these CBF changes.

Distinct brain morphometry patterns revealed by deep learning improve prediction of aphasia severity

Emerging evidence suggests that post-stroke aphasia severity depends on the integrity of the brain beyond the stroke lesion. While measures of lesion anatomy and brain integrity combine synergistically to explain aphasic symptoms, significant interindividual variability remains unaccounted for. A possible explanatory factor may be the spatial distribution of brain atrophy beyond the lesion. This includes not just the specific brain areas showing atrophy, but also distinct three-dimensional patterns of atrophy. Here, we tested whether deep learning with Convolutional Neural Networks (CNN) on whole brain morphometry (i.e., segmented tissue volumes) and lesion anatomy can better predict which individuals with chronic stroke (N=231) have severe aphasia, and whether encoding spatial dependencies in the data might be capable of improving predictions by identifying unique individualized spatial patterns. We observed that CNN achieves significantly higher accuracy and F1 scores than Support Vector Machine (SVM), even when the SVM is nonlinear or integrates linear and nonlinear dimensionality reduction techniques. Performance parity was only achieved when the SVM was directly trained on the latent features learned by the CNN. Saliency maps demonstrated that the CNN leveraged widely distributed patterns of brain atrophy predictive of aphasia severity, whereas the SVM focused almost exclusively on the area around the lesion. Ensemble clustering of CNN saliency maps revealed distinct morphometry patterns that were unrelated to lesion size, highly consistent across individuals, and implicated unique brain networks associated with different cognitive processes as measured by the wider neuroimaging literature. Individualized predictions of severity depended on both ipsilateral and contralateral features outside of the location of stroke. Our findings illustrate that three-dimensional network distributions of atrophy in individuals with aphasia are directly associated with aphasia severity, underscoring the potential for deep learning to improve prognostication of behavioral outcomes from neuroimaging data, and highlighting the prospective benefits of interrogating spatial dependence at different scales in multivariate feature space.

Understanding recovery of language after stroke: insights from neurovascular MRI studies

Stroke causes a disruption in blood flow to the brain that can lead to profound language impairments. Understanding the mechanisms of language recovery after stroke is crucial for the prognosis and effective rehabilitation of people with aphasia. While the role of injured brain structures and disruptions in functional connectivity have been extensively explored, the relationship between neurovascular measures and language recovery in both early and later stages has not received sufficient attention in the field. Fully functioning healthy brain tissue requires oxygen and nutrients to be delivered promptly via its blood supply. Persistent decreases in blood flow after a stroke to the remaining non-lesioned tissue have been shown to contribute to poor language recovery. The goal of the current paper is to critically examine stroke studies looking at the relationship between different neurovascular measures and language deficits and mechanisms of language recovery via changes in neurovascular metrics. Measures of perfusion or cerebral blood flow (CBF) and cerebrovascular reactivity (CVR) provide complementary approaches to understanding neurovascular mechanisms post stroke by capturing both cerebral metabolic demands and mechanical vascular properties. While CBF measures indicate the amount of blood delivered to a certain region and serve as a proxy for metabolic demands of that area, CVR indices reflect the ability of the vasculature to recruit blood flow in response to a shortage of oxygen, such as when one is holding their breath. Increases in CBF during recovery beyond the site of the lesion have been shown to promote language gains. Similarly, CVR changes, when collateral vessels are recruited to help reorganize the flow of blood in hypoperfused regions, have been related to functional recovery post stroke. In the current review, we highlight the main findings in the literature investigating neurovascular changes in stroke recovery with a particular emphasis on how language abilities can be affected by changes in CBF and CVR. We conclude by summarizing existing methodological challenges and knowledge gaps that need to be addressed in future work in this area, outlining a promising avenue of research.

Comparing the brain-behaviour relationship in acute and chronic stroke aphasia

In stroke aphasia, lesion volume is typically associated with aphasia severity. Although this relationship is likely present throughout recovery, different factors may affect lesion volume and behaviour early into recovery (acute) and in the later stages of recovery (chronic). Therefore, studies typically separate patients into two groups (acute/chronic), and this is often accompanied with arguments for and against using data from acute stroke patients over chronic. However, no comprehensive studies have provided strong evidence of whether the lesion-behaviour relationship early in recovery is comparable to later in the recovery trajectory. To that end, we investigated two aims: (i) whether lesion data from acute and chronic patients yield similar results in region-based lesion-symptom mapping analyses and (ii) if models based on one timepoint accurately predict the other. Lesions and aphasia severity scores from acute (N = 63) and chronic (N = 109) stroke survivors with aphasia were entered into separate univariate region-based lesion-symptom mapping analyses. A support vector regression model was trained on lesion data from either the acute or chronic data set to give an estimate of aphasia severity. Four model-based analyses were conducted: trained on acute/chronic using leave-one-out, tested on left-out behaviour or trained on acute/chronic to predict the other timepoint. Region-based lesion-symptom mapping analyses identified similar but not identical regions in both timepoints. All four models revealed positive correlations between actual and predicted Western Aphasia Battery-Revised aphasia-quotient scores. Lesion-to-behaviour predictions were almost equivalent when comparing within versus across stroke stage, despite differing lesion size/locations and distributions of aphasia severity between stroke timepoints. This suggests that research investigating the brain-behaviour relationship including subsets of patients from only one timepoint may also be applicable at other timepoints, although it is important to note that these comparable findings may only be seen using broad measures such as aphasia severity, rather than those aimed at identifying more specific deficits. Subtle differences found between timepoints may also be useful in understanding the nature of lesion volume and aphasia severity over time. Stronger correlations found when predicting acute behaviour (e.g. predicting acute: r = 0.6888, P < 0.001, predicting chronic r = 0.5014, P < 0.001) suggest that the acute lesion/perfusion patterns more accurately capture the critical changes in underlying vascular territories. Differences in critical brain regions between timepoints may shed light on recovery patterns. Future studies could focus on a longitudinal design to compare acute and chronic patients in a more controlled manner.

Predicting Outcomes of Language Rehabilitation: Prognostic Factors for Immediate and Long-Term Outcomes After Aphasia Therapy

BACKGROUND

Aphasia therapy is an effective approach to improve language function in chronic aphasia. However, it remains unclear what prognostic factors facilitate therapy response at the individual level. Here, we utilized data from the POLAR (Predicting Outcomes of Language Rehabilitation in Aphasia) trial to (a) determine therapy-induced change in confrontation naming and long-term maintenance of naming gains and (b) examine the extent to which aphasia severity, age, education, time postonset, and cognitive reserve predict naming gains at 1 week, 1 month, and 6 months posttherapy.

METHOD

A total of 107 participants with chronic (≥ 12 months poststroke) aphasia underwent extensive case history, cognitive-linguistic testing, and a neuroimaging workup prior to receiving 6 weeks of impairment-based language therapy. Therapy-induced change in naming performance (measured as raw change on the 175-item Philadelphia Naming Test [PNT]) was assessed 1 week after therapy and at follow-up time points 1 month and 6 months after therapy completion. Change in naming performance over time was evaluated using paired t tests, and linear mixed-effects models were constructed to examine the association between prognostic factors and therapy outcomes.

RESULTS

Naming performance was improved by 5.9 PNT items (Cohen's d = 0.56, p < .001) 1 week after therapy and by 6.4 (d = 0.66, p < .001) and 7.5 (d = 0.65, p < .001) PNT items at 1 month and 6 months after therapy completion, respectively. Aphasia severity emerged as the strongest predictor of naming improvement recovery across time points; mild (ß = 5.85-9.02) and moderate (ß = 9.65-11.54) impairment predicted better recovery than severe (ß = 1.31-3.37) and very severe (ß = 0.20-0.32) aphasia. Age was an emergent prognostic factor for recovery 1 month (ß = -0.14) and 6 months (ß = -0.20) after therapy, and time postonset (ß = -0.05) was associated with retention of naming gains at 6 months posttherapy.

CONCLUSIONS

These results suggest that therapy-induced naming improvement is predictable based on several easily measurable prognostic factors. Broadly speaking, these results suggest that prognostication procedures in aphasia therapy can be improved and indicate that personalization of therapy is a realistic goal in the near future.

SUPPLEMENTAL MATERIAL

https://doi.org/10.23641/asha.22141829.

A Comparative Investigation of Automatic Speech Recognition Platforms for Aphasia Assessment Batteries

The rehabilitation of aphasics is fundamentally based on the assessment of speech impairment. Developing methods for assessing speech impairment automatically is important due to the growing number of stroke cases each year. Traditionally, aphasia is assessed manually using one of the well-known assessment batteries, such as the Western Aphasia Battery (WAB), the Chinese Rehabilitation Research Center Aphasia Examination (CRRCAE), and the Boston Diagnostic Aphasia Examination (BDAE). In aphasia testing, a speech-language pathologist (SLP) administers multiple subtests to assess people with aphasia (PWA). The traditional assessment is a resource-intensive process that requires the presence of an SLP. Thus, automating the assessment of aphasia is essential. This paper evaluated and compared custom machine learning (ML) speech recognition algorithms against off-the-shelf platforms using healthy and aphasic speech datasets on the naming and repetition subtests of the aphasia battery. Convolutional neural networks (CNN) and linear discriminant analysis (LDA) are the customized ML algorithms, while Microsoft Azure and Google speech recognition are off-the-shelf platforms. The results of this study demonstrated that CNN-based speech recognition algorithms outperform LDA and off-the-shelf platforms. The ResNet-50 architecture of CNN yielded an accuracy of 99.64 ± 0.26% on the healthy dataset. Even though Microsoft Azure was not trained on the same healthy dataset, it still generated comparable results to the LDA and superior results to Google's speech recognition platform.

Mapping lesion, structural disconnection, and functional disconnection to symptoms in semantic aphasia

Patients with semantic aphasia have impaired control of semantic retrieval, often accompanied by executive dysfunction following left hemisphere stroke. Many but not all of these patients have damage to the left inferior frontal gyrus, important for semantic and cognitive control. Yet semantic and cognitive control networks are highly distributed, including posterior as well as anterior components. Accordingly, semantic aphasia might not only reflect local damage but also white matter structural and functional disconnection. Here, we characterise the lesions and predicted patterns of structural and functional disconnection in individuals with semantic aphasia and relate these effects to semantic and executive impairment. Impaired semantic cognition was associated with infarction in distributed left-hemisphere regions, including in the left anterior inferior frontal and posterior temporal cortex. Lesions were associated with executive dysfunction within a set of adjacent but distinct left frontoparietal clusters. Performance on executive tasks was also associated with interhemispheric structural disconnection across the corpus callosum. In contrast, poor semantic cognition was associated with small left-lateralized structurally disconnected clusters, including in the left posterior temporal cortex. Little insight was gained from functional disconnection symptom mapping. These results demonstrate that while left-lateralized semantic and executive control regions are often damaged together in stroke aphasia, these deficits are associated with distinct patterns of structural disconnection, consistent with the bilateral nature of executive control and the left-lateralized yet distributed semantic control network.

Brain age predicts long-term recovery in post-stroke aphasia

The association between age and language recovery in stroke remains unclear. Here, we used neuroimaging data to estimate brain age, a measure of structural integrity, and examined the extent to which brain age at stroke onset is associated with (i) cross-sectional language performance, and (ii) longitudinal recovery of language function, beyond chronological age alone. A total of 49 participants (age: 65.2 ± 12.2 years, 25 female) underwent routine clinical neuroimaging (T1) and a bedside evaluation of language performance (Bedside Evaluation Screening Test-2) at onset of left hemisphere stroke. Brain age was estimated from enantiomorphically reconstructed brain scans using a machine learning algorithm trained on a large sample of healthy adults. A subsample of 30 participants returned for follow-up language assessments at least 2 years after stroke onset. To account for variability in age at stroke, we calculated proportional brain age difference, i.e. the proportional difference between brain age and chronological age. Multiple regression models were constructed to test the effects of proportional brain age difference on language outcomes. Lesion volume and chronological age were included as covariates in all models. Accelerated brain age compared with age was associated with worse overall aphasia severity (F(1, 48) = 5.65, P = 0.022), naming (F(1, 48) = 5.13, P = 0.028), and speech repetition (F(1, 48) = 8.49, P = 0.006) at stroke onset. Follow-up assessments were carried out ≥2 years after onset; decelerated brain age relative to age was significantly associated with reduced overall aphasia severity (F(1, 26) = 5.45, P = 0.028) and marginally failed to reach statistical significance for auditory comprehension (F(1, 26) = 2.87, P = 0.103). Proportional brain age difference was not found to be associated with changes in naming (F(1, 26) = 0.23, P = 0.880) and speech repetition (F(1, 26) = 0.00, P = 0.978). Chronological age was only associated with naming performance at stroke onset (F(1, 48) = 4.18, P = 0.047). These results indicate that brain age as estimated based on routine clinical brain scans may be a strong biomarker for language function and recovery after stroke.

Edge-centric analysis of stroke patients: An alternative approach for biomarkers of lesion recovery

Most neuroimaging studies of post-stroke recovery rely on analyses derived from standard node-centric functional connectivity to map the distributed effects in stroke patients. Here, given the importance of nonlocal and diffuse damage, we use an edge-centric approach to functional connectivity in order to provide an alternative description of the effects of this disorder. These techniques allow for the rendering of metrics such as normalized entropy, which describes the diversity of edge communities at each node. Moreover, the approach enables the identification of high amplitude co-fluctuations in fMRI time series. We found that normalized entropy is associated with stroke lesion severity and continually increases across the time of patients' recovery. Furthermore, high amplitude co-fluctuations not only relate to the lesion severity but are also associated with patients' level of recovery. The current study is the first edge-centric application for a clinical population in a longitudinal dataset and demonstrates how a different perspective for functional data analysis can further characterize topographic modulations of brain dynamics.

Multimodal Neural and Behavioral Data Predict Response to Rehabilitation in Chronic Poststroke Aphasia

BACKGROUND

Poststroke recovery depends on multiple factors and varies greatly across individuals. Using machine learning models, this study investigated the independent and complementary prognostic role of different patient-related factors in predicting response to language rehabilitation after a stroke.

METHODS

Fifty-five individuals with chronic poststroke aphasia underwent a battery of standardized assessments and structural and functional magnetic resonance imaging scans, and received 12 weeks of language treatment. Support vector machine and random forest models were constructed to predict responsiveness to treatment using pretreatment behavioral, demographic, and structural and functional neuroimaging data.

RESULTS

The best prediction performance was achieved by a support vector machine model trained on aphasia severity, demographics, measures of anatomic integrity and resting-state functional connectivity (F1=0.94). This model resulted in a significantly superior prediction performance compared with support vector machine models trained on all feature sets (F1=0.82, P<0.001) or a single feature set (F1 range=0.68-0.84, P<0.001). Across random forest models, training on resting-state functional magnetic resonance imaging connectivity data yielded the best F1 score (F1=0.87).

CONCLUSIONS

While behavioral, multimodal neuroimaging data and demographic information carry complementary information in predicting response to rehabilitation in chronic poststroke aphasia, functional connectivity of the brain at rest after stroke is a particularly important predictor of responsiveness to treatment, both alone and combined with other patient-related factors.

Multi-model mapping of phonemic fluency

The voluntary generation of non-overlearned responses is usually assessed with phonemic fluency. Like most frontal tasks, it draws upon different complex processes and systems whose precise nature is still incompletely understood. Many claimed aspects regarding the pattern of phonemic fluency performance and its underlying anatomy remain controversial. Major limitations of past investigations include small sample size, scant analysis of phonemic output and methodologically insufficient lesion analysis approaches. We investigated a large number of patients with focal unilateral right or left frontal (n = 110) or posterior (n = 100) or subcortical (n = 65) lesions imaged with magnetic resonance or computed tomography and compared their performance on the number of overall responses, words produced over time, extremely infrequent/unknown words and inappropriate words generated. We also employed, for the first time parcel-based lesion-symptom mapping, tract-wise statistical analysis as well as Bayesian multi-variate analysis based on meta-analytically defined functional region of interest, including their interactions. We found that left frontal damage was associated with greater impairment than right frontal or posterior damage on overall fluency performance, suggesting that phonemic fluency shows specificity to frontal lesions. We also found that subcorticals, similar to frontals, performed significantly worse than posteriors on overall performance suggesting that subcortical regions are also involved. However, only frontal effects were found for words produced over time, extremely infrequent/unknown and inappropriate words. Parcel-based lesion-symptom mapping analysis found that worse fluency performance was associated with damage to the posterior segment of the left frontal middle and superior gyrus, the left dorsal anterior cingulate gyrus and caudate nucleus. Tract-wise statistical analysis revealed that disconnections of left frontal tracts are critical. Bayesian multi-variate models of lesions and disconnectome maps implicated left middle and inferior frontal and left dorsomedial frontal regions. Our study suggests that a set of well localized left frontal areas together with subcortical regions and several left frontal tracts are critical for word generation. We speculate that a left lateralized network exists. It involves medial, frontal regions supporting the process of 'energization', which sustains activation for the duration of the task and middle and inferior frontal regions concerned with 'selection', required due to the competition produced by associated stored words, respectively. The methodology adopted represents a promising and empirically robust approach in furthering our understanding of the neurocognitive architecture underpinning executive processes.

Machine learning-based multimodal prediction of language outcomes in chronic aphasia

Recent studies have combined multiple neuroimaging modalities to gain further understanding of the neurobiological substrates of aphasia. Following this line of work, the current study uses machine learning approaches to predict aphasia severity and specific language measures based on a multimodal neuroimaging dataset. A total of 116 individuals with chronic left-hemisphere stroke were included in the study. Neuroimaging data included task-based functional magnetic resonance imaging (fMRI), diffusion-based fractional anisotropy (FA)-values, cerebral blood flow (CBF), and lesion-load data. The Western Aphasia Battery was used to measure aphasia severity and specific language functions. As a primary analysis, we constructed support vector regression (SVR) models predicting language measures based on (i) each neuroimaging modality separately, (ii) lesion volume alone, and (iii) a combination of all modalities. Prediction accuracy across models was subsequently statistically compared. Prediction accuracy across modalities and language measures varied substantially (predicted vs. empirical correlation range: r = .00-.67). The multimodal prediction model yielded the most accurate prediction in all cases (r = .53-.67). Statistical superiority in favor of the multimodal model was achieved in 28/30 model comparisons (p-value range: <.001-.046). Our results indicate that different neuroimaging modalities carry complementary information that can be integrated to more accurately depict how brain damage and remaining functionality of intact brain tissue translate into language function in aphasia.

Performance Evaluation of Machine Learning Frameworks for Aphasia Assessment

Speech assessment is an essential part of the rehabilitation procedure for patients with aphasia (PWA). It is a comprehensive and time-consuming process that aims to discriminate between healthy individuals and aphasic patients, determine the type of aphasia syndrome, and determine the patients' impairment severity levels (these are referred to here as aphasia assessment tasks). Hence, the automation of aphasia assessment tasks is essential. In this study, the performance of three automatic speech assessment models based on the speech dataset-type was investigated. Three types of datasets were used: healthy subjects' dataset, aphasic patients' dataset, and a combination of healthy and aphasic datasets. Two machine learning (ML)-based frameworks, classical machine learning (CML) and deep neural network (DNN), were considered in the design of the proposed speech assessment models. In this paper, the DNN-based framework was based on a convolutional neural network (CNN). Direct or indirect transformation of these models to achieve the aphasia assessment tasks was investigated. Comparative performance results for each of the speech assessment models showed that quadrature-based high-resolution time-frequency images with a CNN framework outperformed all the CML frameworks over the three dataset-types. The CNN-based framework reported an accuracy of 99.23 ± 0.003% with the healthy individuals' dataset and 67.78 ± 0.047% with the aphasic patients' dataset. Moreover, direct or transformed relationships between the proposed speech assessment models and the aphasia assessment tasks are attainable, given a suitable dataset-type, a reasonably sized dataset, and appropriate decision logic in the ML framework.