Cathy Price

Post-stroke aphasia is a major contributor to language impairment and neuro-disability worldwide, making automated assessment a critical research priority. However, clinically validated automatic speech recognition (ASR) systems remain limited by the scarcity of large, annotated datasets capturing aphasia’s heterogeneous manifestations. We introduce SONIVA (Speech recOgNItion Validation in Aphasia), the largest and most comprehensively curated database for validating speech recognition in aphasia, comprising audio recordings from approximately 1,000 stroke survivors and 6,000 age-matched controls. The dataset comprises annotated speech from 571 stroke survivors, 103 of whom contributed longitudinal recordings (mean age: 60.65 ± 12.97 years; 68.77% male), and 103 controls (mean age: 59.64 ± 11.48 years; 62.05% male). These recordings are enriched with detailed linguistic coding, orthographic transcriptions, and International Phonetic Alphabet annotations. Foundation models fine-tuned on SONIVA correlate strongly with expert transcriptions (Spearman’s r = 0.79-0.86; p < 0.0001), while acoustic classifiers achieve 93% stroke classification accuracy, enabling scalable analysis for rehabilitation and clinical assessment.

Aims: A qualitative process evaluation was embedded into the pilot phase of the RADICAL randomised controlled trial of radiofrequency denervation for chronic and moderate-severe low back pain. The aim was to rapidly understand recruitment processes and inform development of strategies to overcome recruitment challenges. Methods: Audio recording of recruitment consultations between recruiting staff and potential patient participants, and interviews with patients who had consented to participate in the RADICAL trial. We used rapid thematic analysis to identify what information was provided, understanding of trial processes and potential challenges to recruitment. Results: Recruitment staff needed more support to clearly and effectively explain trial participation. Specific challenges included communicating trial procedures, explaining treatment equipoise and randomisation, and navigating the complex recruitment pathway. These findings led to the development of support materials for recruiters, including a process diagram and targeted training sessions to optimise information provision and informed consent. Conclusion: Embedding qualitative process evaluations within an RCT pilot can generate rapid insights into communication challenges and recruitment barriers. These insights can inform the development of practical support materials and process refinements to enhance recruitment, potentially strengthening trial viability.

Functional imaging and clinical cases implicate the left thalamus in object naming, yet the prevalence of naming impairment after focal thalamic damage is low with variable impact and often rapid resolution. This suggests that compensatory mechanisms, within or beyond the thalamus, may support recovery. We hypothesized that thalamic damage would (a) not cause chronic anomia if other naming-related regions remain intact but (b) exacerbate anomia when co-occurring with damage to non-thalamic naming regions. To test these hypotheses, we retrospectively assessed naming ability in 550 left hemisphere chronic stroke survivors (52% with anomia). Lesion sites included focal thalamic lesions (n = 14), combined thalamic and non-thalamic lesions (n = 271), and lesions sparing the thalamus (n = 265). Whole-brain lesion–symptom mapping (LSM), using multivariate support vector regression, identified brain regions where damage was significantly related to naming ability. Contributions of different thalamic subregions to naming were assessed using ridge regression. Focal thalamic lesions were not associated with chronic anomia. LSM identified two naming-related clusters: a temporoparietal region of interest (ROI-TP) and a subcortical–insular region of interest (ROI-SC) including the lateral thalamus. However, lesion load in the lateral thalamus did not independently contribute to naming performance when controlling for damage to other parts of the ROI-SC, nor did any thalamic nuclei show additive effects beyond the ROI-TP and the non-thalamic ROI-SC. These findings suggest that thalamic damage in the dominant hemisphere does not cause long-term anomia in chronic stroke. Future research therefore needs longitudinal designs to track the trajectory of transient thalamic effects from the acute to chronic phases and to investigate whether naming impairments after thalamic lesions are (a) lesion specific but context dependent, emerging under increased cognitive load, or (b) attributable to non-lesion-site-dependent post-stroke factors such as fatigue.

Introduction: This study investigates how chronological age at stroke onset affects post‐stroke language abilities and recovery. Using cross‐sectional and retrospective data, we investigated (A) which language tasks are sensitive to age; (B) whether any age‐related effects are modified by factors such as lesion size, initial severity and education and (C) whether age influences recovery from aphasia. Methods: Language abilities were assessed in 749 participants using (i) the Comprehensive Aphasia Test (CAT) at a single time point and (ii) self‐reported measures of speaking, understanding, reading and writing ability both at 1 week and 1 year post‐stroke. The outcomes investigated, using multiple linear regression, were (1) CAT language scores and (2) self‐reported recovery scores at 1 year post‐stroke. Lesion size, initial severity, sex, handedness, pre‐stroke education and time post‐stroke were controlled. Bayesian statistics assessed the strength of evidence for observed effects. Results: (A) Irrespective of lesion or symptom severity, older age at stroke onset was associated with worse performance across multiple language domains, including overall language ability, nonword and word repetition, semantic and letter fluency, object naming and comprehension of spoken and written words and sentences. Effect sizes were strongest for nonword repetition ( β = −0.258) and weakest for written word comprehension ( β = −0.061). (B) The disadvantage of older age was heightened in participants with the largest lesions and most severe initial aphasia, but (C) there was no evidence that older age hindered language recovery in any subgroup. Conclusion: Older age at stroke onset significantly worsened language outcomes, highlighting the sensitivity of aphasia assessments to age‐related differences in post‐stroke language abilities. This disadvantage was greatest in participants with both large lesions and severe initial symptoms. Considering age when interpreting assessment scores would optimise diagnosis and prognostic accuracy.

Reading is central to academic and vocational success. Some deaf children face reading challenges due to limited access to spoken or signed language. Robust phonological representations are key to reading development in hearing children. Spoken language phonology may be one of many contributors to reading development in deaf children. Indeed, speechreading ability correlates with reading skill in both deaf and hearing individuals, suggesting it is linked to reading development regardless of hearing status. Further support for this hypothesis would be provided by evidence that similar neural representations of speech phonology are evoked by visual speech and other language forms (auditory speech and text), and that these neural representations are related to reading proficiency. We used fMRI and Representational Similarity Analysis (RSA) to identify shared neural representations of spoken language phonological structure. A group of deaf adult participants (N=22), with a mixture of sign language and spoken language backgrounds and reading abilities, were presented with single lexical items as visual speech and dynamic text (cursive text, revealed letter-by-letter to promote a phonological reading strategy). Adult hearing participants (N=25) were presented with the same words, but as visual speech and auditory speech. Shared neural representations of phonological structure of English words were found in each group in the superior and middle temporal cortex (STC/MTC) and these abstract representations were more similar across different language forms in better readers. Our data provide neurobiological evidence of the contribution of visual speech to abstract phonological representations of spoken language, that relate to reading proficiency, in both deaf and hearing adults. Significance Statement: Reading is an essential skill, yet some deaf children face reading challenges due to reduced access to signed or spoken language. In hearing children, successful reading depends on abstract phonological representations, but whether spoken language phonology relates to reading in deaf individuals remains unclear. Using fMRI and RSA, we show that deaf and hearing adults recruit neural representations of phonology that are shared by visual speech and other language forms (visual/auditory speech in hearing; visual speech/dynamic text in deaf) in the superior and middle temporal cortex. Critically, greater cross-modal alignment of neural representations of phonological structure was associated with better reading in both groups. These findings provide neurobiological evidence that visual speech contributes to phonological representations that relate to reading, regardless of hearing status.

Current deep learning-based approaches to lesion segmentation in neuroimaging often depend on high-resolution images and extensive annotated data, limiting clinical applicability. This paper introduces a novel synthetic data framework tailored for stroke lesion segmentation, expanding the SynthSeg methodology to incorporate lesion-specific augmentations that simulate diverse pathological features. Using a modified nnUNet architecture, our approach trains models with label maps from healthy and stroke datasets, facilitating segmentation across both normal and pathological tissue without reliance on specific sequence-based training. Our method achieves robust out-of-domain performance where conventional approaches fail, with in-domain performance of 48.2% Dice compared to 57.5% for conventional training. Crucially, even with oracle knowledge of the optimal domain adaptation method - an unrealistic scenario in practice - conventionally-trained models cannot match our synthetic approach in out-of-domain settings. The framework demonstrates that synthetic pre-training provides fundamental robustness unachievable through test-time adaptation alone. Our approach reduces reliance on domain-specific training data and helps bridge the gap between research-grade and clinical scans to improve clinical stroke neuroimaging workflows. PyTorch training code and weights are publicly available at https://github.com/liamchalcroft/SynthStroke, along with an SPM toolbox featuring a plug-and-play model at https://github.com/liamchalcroft/SynthStrokeSPM

Post-stroke aphasia is a major contributor to language impairment and neuro-disability worldwide, making automated assessment a critical research priority. However, the development of clinically validated automatic speech recognition (ASR) systems remains limited by the lack of large, annotated datasets that capture aphasia’s heterogeneous and unpredictable manifestations. We introduce SONIVA (Speech recOgNItion Validation in Aphasia), the largest and most richly annotated database of pathological speech to date, comprising recordings from ≈1,000 stroke survivors (including over 200 longitudinally) and ≈7,000 age-matched controls. Current annotations include 576 patients (mean age: 61.23 ± 13.23 years; 69.81% male) and 104 controls (mean age: 61.05 ± 12.05 years; 34% male), with rich linguistic coding, orthographic and international phonetic alphabet transcriptions. Foundation models finetuned on SONIVA extract linguistic features that correlate with expert transcriptions (Spearman’s r = 0.86 - 0.79; p < 0.0001), while acoustic classifiers achieve 93% stroke classification accuracy. These results position SONIVA as a critical resource that can transform rehabilitation through objective, scalable speech assessment.

Ample reports highlight fMRI’s added value to guide neurosurgical interventions near brain regions supporting speech and language. However, fMRI’s usefulness for clinical language mapping remains controversial, partly fueled by 1) differences from clinical standard tools it is often compared against, and 2) wide heterogeneity in how data are acquired, analyzed and interpreted. Both factors limit objective assessment of the benefits and efficacy of presurgical fMRI. This underscores the need for standardization of fMRI protocols to enable data pooling across centers and facilitate learning from patient outcomes. The OHBM Working Group on clinical fMRI language mapping was formed in 2017. Its scope was to review and propose best practice recommendations addressing specific challenges posed by applications in patient populations. Objectives were to: 1) consider language tasks and designs, optimized for specific clinical objectives, and incorporating modifications for patients with existing impairments; 2) offer practical guidance, based on high-quality research, for each step from fMRI acquisition and analysis to reporting individual patients’ data. In considering these challenges we focus on implementations that have proven feasible based on approaches in active use today. When widely available practices deviate from optimal practices, we highlight emerging developments meriting further evaluation and incorporation into clinical use. This document was created in collaboration with the OHBM Committee on Best Practices, incorporating community feedback. It aims to provide a framework for improved standardization of fMRI to enable much-needed evaluations of its ultimate goals; namely, minimization of invasive intraoperative testing and, ultimately, of new post-operative language deficits. Accordingly, the single strongest recommendation is for greater transparency and reporting of longitudinal outcomes in patients undergoing clinical fMRI.