Back to search

Cumming School of Medicine

Automatic Perfusion Imaging Analysis for the Prediction of Tissue Fate in Acute Ischemic Stroke using Machine Learning

Abstract

dc:description.abstract

In the domain of acute ischemic stroke, tissue outcome prediction is a computational task uniquely positioned at the intersection of medicine and data science. It is used by clinicians to make acute treatment decisions and by researchers to select patients for clinical trials. Machine learning methods have the potential to improve the process of stroke tissue outcome prediction by utilizing the full multi-modal patient datasets typically acquired in the clinical routine. However, technical details related to data preprocessing and machine learning classifier design vary widely between studies, making it difficult to reach a consensus on how best to apply these machine learning-based methods to real-life medical scenarios. The research described in this thesis aims to establish such a consensus through comparative studies of classical machine learning and deep learning experimental designs. To address the wide range of experimental designs proposed for stroke tissue outcome prediction using classical machine learning methods, this work exhaustively compared common design choices from the published literature to establish a top-performing ’baseline’ model design. To address inconsistencies in the data processing techniques used to prepare imaging features for stroke tissue outcome prediction, this work also proposed a method to automatically select the arterial input function of computed tomography perfusion and perfusion-weighted MRI datasets. Using the baseline classical machine learning model design and proposed data preprocessing method, stroke tissue outcome predictions were generated to perform the clinically relevant task of in-silico stroke treatment efficacy evaluation. It was shown that the proposed in-silico study design can outperform conventional treatment efficacy analysis for a small patient population. Additionally, the established classical machine learning design was compared to a novel deep learning classifier, which revealed that deep learning can significantly outperform classical machine learning for stroke tissue outcome prediction. Overall, this work provides a clear and detailed framework for training classical machine learning models that are clinically relevant to stoke treatment efficacy analysis, and further suggests that the use of novel deep learning methods may constitute an improvement over classical machine learning in future studies.

Degree

thesis:*
Name thesis:degree_name
Master of Science (MSc)
Discipline thesis:degree_discipline
Medicine – Neuroscience
Grantor
Cumming School of Medicine
Year dc:date.issued
2022

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Winder, Anthony J
Advisor dc:contributor.advisor
  • Forkert, Nils D
Committee members dc:contributor.committeemember
  • Forkert, Nils D
  • Menon, Bijoy K
  • Lee, Ting-Yim

Subjects

dc:subject × 13

Rights

Language dc:language.iso
English

Identifiers

dc:identifier.*
Dc Identifier Citation
Winder, A. J. (2022). Automatic Perfusion Imaging Analysis for the Prediction of Tissue Fate in Acute Ischemic Stroke using Machine Learning (Master thesis). University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca .

Chain of custody

source
Harvested from
University of Calgary
Base URL
ucalgary.scholaris.ca/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Winder, Anthony J. Automatic Perfusion Imaging Analysis for the Prediction of Tissue Fate in Acute Ischemic Stroke using Machine Learning. Cumming School of Medicine, 2022. http://hdl.handle.net/1880/116309