Prognostication of large-vessel ischaemic stroke treated with endovascular thrombectomy
Date
2024
Authors
Zeng, Minyan
Editors
Advisors
Palmer, Lyle
Jenkinson, Mark
Jannes, Jim
Jenkinson, Mark
Jannes, Jim
Journal Title
Journal ISSN
Volume Title
Type:
Thesis
Citation
Statement of Responsibility
Conference Name
Abstract
Prognostic prediction for large vessel occlusion stroke is critical for effective care planning and patient management. Prognostic prediction at admission informs treatment strategies, optimises resource allocation, and helps set realistic expectations for patients and their families. This thesis focuses on improving prognostic prediction for large vessel occlusion stroke using comprehensive clinical data and imaging modalities. First, a study used multivariable regression models to predict futile recanalisation, 3-month functional outcomes and long-term mortality based on routinely collected hospital admission data. These models showed satisfactory predictive performance, supporting informed treatment decisions and patient-informed consent in clinical practice. Building on this, a systematic review was conducted to evaluate existing prognostic models using more advanced techniques – machine learning and deep learning techniques. This review highlighted the potential of these advanced methods and indicated the need for their application to larger and more diverse datasets to enhance model performance. Thus, exploratory experiments were initially conducted to investigate the use of multimodal imaging data, including NCCT, CTA, and CTP maps, in model prediction. Several innovative strategies were explored, with incorporating CTP maps and using an ensemble strategy in the model showing potential for improved performance. Further research focused on incorporating the benefits of CTP maps into full modalities models and explored the possibility of synthesising these maps from NCCT and CTA images. The approach demonstrated that network architectures incorporating CTP maps for prognostic prediction can be flexibly deployed when the maps are unavailable, as their benefits can be effectively synthesised from NCCT and CTA. Finally, the thesis updated prognostic prediction models using an up-to-date, larger dataset and conducted various analyses to assess their performance across different patient subgroups and clinical scenarios. The results showed a marginal improvement in the predictive ability of models when imaging was incorporated and the influence of patient age on model performance was identified. In summary, this thesis advances the field of LVO stroke prognostic prediction by integrating comprehensive data and leveraging traditional statistical modelling, machine learning, and deep learning techniques. The findings show promise in enhancing prognostic predictions of LVO stroke using deep learning techniques but highlight the need for technological innovation and multicentre collaboration in data collection to continually improve the prediction models.
School/Discipline
School of Public Health
Dissertation Note
Thesis (Ph.D.) -- University of Adelaide, School of Public Health, 2024
Provenance
This electronic version is made publicly available by the University of Adelaide in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. This thesis may incorporate third party material which has been used by the author pursuant to Fair Dealing exceptions. If you are the owner of any included third party copyright material you wish to be removed from this electronic version, please complete the take down form located at: http://www.adelaide.edu.au/legals