Electronic Thesis and Dissertation Repository

Thesis Format

Monograph

Degree

Master of Science

Program

Neuroscience

Supervisor

Owen, Adrian M.

2nd Supervisor

Debicki, Derek B.

Co-Supervisor

Abstract

Assessment of consciousness in behaviourally unresponsive patients with critical brain injuries continues to be a challenge. There remains a need for robust tools that can accurately characterize preserved cortical function and predict patient outcomes. In the present study, functional near-infrared spectroscopy is employed in conjunction with graph theory and machine learning to quantify resting-state functional connectivity in 16 acutely brain-injured patients and 23 healthy controls. Results revealed significant channel-level differences between the groups for three graph metrics, including degree, clustering coefficient, and local efficiency. Further investigation using machine learning algorithms revealed that these metrics can be used to distinguish between patients and healthy controls with 76% accuracy, and between good and poor patient outcomes with 83% accuracy. Overall, findings from this study provide valuable insights into alterations in brain connectivity following acute brain injury, along with a robust statistical approach for determining patient diagnosis and prognosis.

Summary for Lay Audience

Assessing consciousness in unresponsive patients with critical brain injuries in the intensive care unit poses a significant challenge for medical professionals. Accurately detecting preserved brain function and predicting patient outcomes is crucial for making informed decisions regarding patient care and management. To tackle these challenges, researchers are exploring the use of advanced functional neuroimaging techniques.

In this study, we employed a mathematical approach called graph theory to analyze patterns of brain connectivity in 16 patients with critical brain injuries and 23 healthy controls. We used a non-invasive brain imaging technique called functional near-infrared spectroscopy to measure brain activity in both groups. Our analyses revealed significant connectivity differences between the patient and healthy control participants. We then used these connectivity measures to develop machine learning algorithms capable of distinguishing between patients and healthy controls with a 76% accuracy rate. Furthermore, the algorithms were able to predict good or poor patient outcomes with an 83% accuracy rate.

The findings of our study contribute to the understanding of how critical brain injuries impact brain connectivity. The results demonstrate the potential for combining advanced brain imaging techniques and mathematical analyses to enhance diagnostic and prognostic precision in this patient population. By developing more accurate and reliable tools to assess brain function in unresponsive patients, we aim to support medical teams and families in making well-informed decisions regarding patient care.

Download

Share

COinS