Publication Date

8-2024

Advisor(s) - Committee Chair

Richard Schugart, Mark Robinson, Lukun Zheng

Degree Program

Department of Mathematics

Degree Type

Master of Science

Abstract

Acute ischemic stroke, caused by cerebral artery blockage, is a leading cause of long-term disability and mortality. Effective management relies on accurate, timely assessments from neuroimaging data. Computed tomography perfusion (CTP) imaging is crucial in evaluating stroke patients, offering detailed maps of cerebral perfusion to identify irreversibly damaged tissue and at-risk areas. This detailed assessment is essential for informed therapeutic decisions.

Key perfusion parameters derived from CTP imaging, including cerebral blood volume (CBV), cerebral blood flow (CBF), time to peak (TTP), and mean transit time (MTT), are crucial for understanding the extent and nature of cerebral ischemia, providing valuable insights into the brain’s vascular physiology and pathophysiology. The advent of deep learning has revolutionized medical imaging, offering powerful tools for automated and accurate image analysis. Within this framework, Temporal Convolutional Neural Networks (TCNNs) have emerged as a promising technique for processing time-varying data, such as CTP imaging datasets.

This thesis explores the application of TCNNs for estimating perfusion parameters from CTP imaging data, demonstrating superior accuracy and robustness compared to CNNs and RNNs. TCNNs provide more precise estimates of CBV, CBF, TTP, and MTT, significantly reducing root mean square error (RMSE), achieving lower final loss, and ensuring rapid convergence across multiple patient datasets. The integration of TCNN-based analysis into clinical workflows is also discussed, highlighting its potential to enhance diagnostic accuracy, inform treatment decisions, and improve patient outcomes in acute ischemic stroke management.

Overall, this thesis underscores the transformative potential of TCNNs in the realm of perfusion imaging. By automating the analysis process and improving the accuracy of perfusion parameter estimation, TCNNs represent a significant advancement in the field of neuroimaging. The findings presented herein pave the way for further research and clinical adoption of deep learning techniques, promising a future where advanced computational tools are seamlessly integrated into the fabric of clinical practice, enhancing the care and management of stroke patients.

Disciplines

Data Science | Medical Sciences | Medical Specialties | Physical Sciences and Mathematics | Radiology

Available for download on Thursday, July 23, 2026

Share

COinS