Thumbnail Image
Person

Hope, Thomas

Collections
JCU Authors
Department of Psychological and Social Sciences
Show all metadata
Institutional profile
Professor Hope began his academic studies at the University of Oxford, reading, thinking and writing about the brain / mind. To try to tackle these issues at a more practical level, he then moved to the University of Sussex to study a sub-field of Artificial Intelligence (AI) that emphasizes the dynamics of sensorimotor integration in embodied agents. Aiming to use these techniques at a more ambitious scale, he took a staff scientist job at the UK’s Defence Evaluation and Research Agency (DERA), developing new AI technology for military applications. And then deciding he needed to better understand the brain, he took up a Marie Curie Early Stage fellowship to study the format of neural semantic representations of number, at the University of Padova (Italy), with stints in Aachen (Germany), and Paris (France). On returning to the UK, Professor Hope worked at Imperial College London, studying ‘clinician resistance’ to the introduction of new AI technology in medicine. He then moved to University College London, to use brain imaging and formal / computational methods to study the neurobiology of language in health and disease. Much of his current research involves developing new applications of machine learning for neurology, and work aiming to elucidate the mechanisms that drive language in the brain. Professor Hope joined John Cabot University in 2022 as a Visiting Professor, and in 2023 as an Associate Professor. Beyond his work in academia and research, he has been a tech entrepreneur and a management consultant, and he still takes occasional work as a technical consultant on AI projects. He's also a trained barrister: a lawyer specializing in oral advocacy in the courts of England and Wales.

Filters

2025 - 20263
Reset filters

Settings

Citations

Publication Search Results

Now showing 1 - 3 of 3
  • Thumbnail Image
    PublicationOpen Access
    Deep convolutional neural networks outperform vanilla machine learning when predicting language outcomes after stroke
    (2025) Hope, Thomas; Bowman, Howard; Leff, Alex P.; Price, Cathy J.
    Background: Current medicine cannot confidently predict patients’ language skills after stroke. In recent years, researchers have sought to bridge this gap with machine learning. These models appear to benefit from access to features describing where and how much brain damage these patients have suffered. Given the very high dimensionality of structural brain imaging data, those brain lesion features are typically post-processed from the images themselves into tabular features. With the introduction of deep Convolutional Neural Networks (CNN), which appear to be much more robust to high dimensional data, it is natural to hope that much of this image post-processing might be unnecessary. But prior attempts to demonstrate this (in the area of post-stroke prognostics) have so far yielded only equivocal results – perhaps because the datasets that those studies could deploy were too small to properly constrain CNNs, which are famously ‘data-hungry’. Methods: The study draws on a much larger dataset than has been employed in previous work like this, referring to patients whose language outcomes were assessed once during the chronic phase post-stroke, on or around the same days as they underwent high resolution MRI brain scans. Following the model of our own and others’ past work, we use state of the art ‘vanilla’ machine learning models (boosted ensembles) to predict a variety of language and cognitive outcomes scores. These models employ both demographic variables and features derived from the brain imaging data, which represent where brain damage has occurred. These are our baseline models. Next, we use deep CNNs to predict the same language scores for the same patients, drawing on both the demographic variables, and post-processed brain lesion images: i.e., multi-input models with one input for tabular features and another for 3-dimensional images. We compare the models using 5 × 2-fold cross-validation, with consistent folds. Results: The CNN models consistently outperform the vanilla machine learning models, in this domain. Conclusions: Deep CNNs offer state of the art performance when predicting language outcomes after stroke, outperforming vanilla machine learning and obviating the need to post-process lesion images into lesion features.
  • Thumbnail Image
    PublicationOpen Access
    Object Naming After Thalamic Damage: Evidence From a Large-Scale, Chronic-Phase Study of Left Hemisphere Stroke Survivors
    (2026) Zhang, Jie; Neville, Douglas; Anderson, Storm; Roberts, Sophie M.; Hope, Thomas; Leff, Alex P.; Green, David W.; Price, Cathy J.
    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.
  • Thumbnail Image
    PublicationOpen Access
    Precision‐Optimised Post‐Stroke Prognoses
    (2025) Hope, Thomas; Bowman, Howard; Bruce, Rachel M.; Leff, Alex P.; Price, Cathy J.
    Background Current medicine cannot confidently predict who will recover from post-stroke impairments. Researchers have sought to bridge this gap by treating the post-stroke prognostic problem as a machine learning problem, reporting prediction error metrics across samples of patients whose outcomes are known. This approach effectively shares prediction error equally among the patients, which is contrary to the long-held clinical intuition that some patients' outcomes are more predictable than other patients' outcomes. Here, we test that intuition empirically, by asking whether those ‘more predictable’ patients can be identified before their outcomes are known. Methods Drawing on lesion location and demographic data, we use ensemble classifiers to predict the presence of a variety of different language impairments in a large sample of stroke patients. We tune these models to maximise their Positive Predictive Value (or precision): that is, the probability that patients assigned to a class are really members of that class. We test whether those tuned models have high precision on independent data. Results Precision-tuned models might only classify a subset of patients, but for that reduced set, the classifications are very likely to be correct: typically > 90% and sometimes > 95%. Small reductions of target precision could rapidly raise the proportion of patients for whom ‘high enough precision’ predictions can be made. Conclusions High precision prognoses are possible when predicting language outcomes after stroke. Providing such predictions for subsets of patients might be a reasonable intermediate step on the way to providing them for all.