Ruffle, James;
(2024)
Graph, deep, and generative models of the human brain in health and disease.
Doctoral thesis (Ph.D), UCL (University College London).
Text
JRuffle_thesis_v1.8_figs_final_redacted.pdf - Accepted Version Access restricted to UCL open access staff until 1 October 2025. Download (259MB) |
Abstract
That the brain exhibits a finely wrought functional, anatomical—high-dimensional—organisation is no longer in doubt. Macro- and micro-structural features, task-specific and resting state neural activity, focal disruptive and lesion-related neural dependence all show richly structured, replicable variation across the population. However, whether these now-familiar brain processing patterns in health aid understanding of differences at the individual level, not least in their application to the diseased brain, remains an open question. The answer to this unknown is vital for two reasons: first, because the clinical applications of our knowledge of the brain are necessarily addressed not to populations but to individual patients, and second, because the fewer the individuals to which any model generalises, the weaker the grounds it provides for mechanistic inference, no matter how well supported its parameters. Understanding the human brain in either health or disease requires a special kind of trinity: large-scale data, high-performance computing, and cutting-edge algorithms, the harmony of which catalyse innovation. In this thesis, we lean on all three to develop a suite of research that advances our understanding of the human brain in health and disease. In health, exploiting recent innovations in Bayesian generative graph models, we advance the understanding of the brain’s autonomic nervous system regulatory function. With deep auto-encoding, we construct a new high-performance means to compress dense transcriptome data across the human brain, applicable to any downstream inferential task. With deep learning of multi-channel volumetric magnetic resonance imaging data, we reveal the computational limits of the imaged human brain in unprecedented detail. In disease, we develop a novel method for lesion deficit mapping that exploits the mathematical principles of the layered Bayesian stochastic block model, marginalising the confounding effects of lesion spatial location from any underlying cognitive task it seeks to uncover: a formal implementation of Ockham’s razor. Finally, we further apply these graph and deep learning tools to neuro-oncology—a group of diseases where treatment and outcome are unchanged for thirty years—to study an unrivalled cohort of patients diagnosed with brain tumours over 15 years in our centre and internationally. In doing so, we derive a new means to stratify patients according to Bayesian graphical modelling of the neuropathology of tumours, with proven superior survival forecasting. We develop a set of tumour segmentation models that are effective regardless of missing MRI sequences, commonplace across clinical practice. Lastly, we harmonise all technical and methodological innovations developed throughout this thesis to undertake deep cartography of human glioma across the imaging, molecular, constitutional, and physiological neural landscape, applied to the task of individualised outcome prediction.
Type: | Thesis (Doctoral) |
---|---|
Qualification: | Ph.D |
Title: | Graph, deep, and generative models of the human brain in health and disease |
Language: | English |
Additional information: | Copyright © The Author [year]. Original content in this thesis is licensed under the terms of the Creative Commons Attribution 4.0 International (CC BY 4.0) Licence (https://creativecommons.org/licenses/by/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. |
UCL classification: | UCL UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Brain Sciences UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Brain Sciences > UCL Queen Square Institute of Neurology UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Brain Sciences > UCL Queen Square Institute of Neurology > Brain Repair and Rehabilitation |
URI: | https://discovery.ucl.ac.uk/id/eprint/10197189 |
1. | United Kingdom | 3 |
Archive Staff Only
View Item |