Abstract
Predicting and supporting language recovery after stroke remains a major clinical challenge. Traditional models based on lesion size and location explain only part of the variability in aphasia outcomes. AI has huge potential to enable more precise, scalable approaches to modelling and enhancing recovery. This talk outlines how multimodal biomarkers, from structural neuroimaging, network-level brain activity to behaviour, can inform individualised recovery predictions. Using digital platforms such as the Imperial Comprehensive Cognitive assessment in Cerebrovascular disease (IC3), longitudinal online testing captures cognitive and language changes remotely and at scale, facilitating better recovery modelling. Other applications of AI-based speech technologies will be discussed, including fine-tuned automatic speech recognition systems retrained on aphasic speech (SONIVA), that significantly improve speech-to-text accuracy and enable automated, objective assessment of language impairment in patients after stroke. Together, these advances support a roadmap toward personalised, data-driven language rehabilitation. Key challenges remain in fairness, interpretability, and clinical integration, but AI promises to transform how we predict, measure, and enhance recovery after brain injury.
Video
