Algorithmic Characterisation of Non-Rigid Registration in Inter-Subject Resting-State Functional Magnetic Resonance Image Processing

Svejda, Martin (2024) Algorithmic Characterisation of Non-Rigid Registration in Inter-Subject Resting-State Functional Magnetic Resonance Image Processing. Doctoral thesis, Birmingham City University.

[thumbnail of Martin Svejda PhD Thesis_Final Version_Final Award Dec 2024.pdf]
Preview
Text
Martin Svejda PhD Thesis_Final Version_Final Award Dec 2024.pdf - Accepted Version

Download (8MB)

Abstract

Resting-State Functional Magnetic Resonance Imaging (rs-fMRI) is fundamental for studying intrinsic brain functions, crucial for defining the networks underlying human cognition and behaviour. Non-rigid registration algorithms are essential for accurately aligning rs-fMRI data across subjects, a process critical for consistent and reliable analysis of functional connectivity. The performance of these algorithms directly impacts the precision of neuroimaging results due to individual anatomical differences.

This thesis addresses the critical issue of performance variability among non-rigid registration algorithms, which can undermine the reliability and accuracy of functional connectivity analysis in rs-fMRI. To systematically assess these differences, the Non-Rigid Registration Algorithm Analysis Framework (NRAAF) was developed and implemented, offering an innovative benchmark for evaluating and characterising the accuracy and specificity of various algorithms.

Key findings show that algorithms such as Advanced Normalisation Tools (ANTs), Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL), Analysis of Functional NeuroImages (AFNI), and FMRIB Software Library (FSL) exhibit significant differences in handling anatomical variability. ANTs demonstrated superior sensitivity with a mean Peak Activation Intensity of 0.85, while DARTEL showed the most consistent performance with minimal variability (Standard Deviation of 0.05). AFNI presented a higher variance in cluster detection at 0.30, compared to FSL’s 0.18. These insights emphasise that algorithm selection crucially influences the reliability of functional connectivity analyses.

The differential performance among these algorithms significantly impacts neuroimaging outcomes, affecting both the interpretation of research findings and potential clinical applications. By providing a comprehensive evaluation and characterisation of non-rigid registration algorithms, this work emphasises the importance of selecting appropriate methods to enhance reproducibility and accuracy in neuroimaging. In doing so, NRAAF framework empowers the neuroimaging community to advance computational methodologies and refine tools for studying complex brain functions, with potential implications for diagnostics, personalised treatment strategies, and broader cross institutional research collaborations.

Item Type: Thesis (Doctoral)
Dates:
Date
Event
19 December 2024
Accepted
Uncontrolled Keywords: Neuroimaging Algorithms, Non-Rigid Registration, Functional Connectivity, Resting-State fMRI Analysis, Multivoxel Pattern Analysis
Subjects: CAH02 - subjects allied to medicine > CAH02-05 - medical sciences > CAH02-05-01 - medical technology
CAH02 - subjects allied to medicine > CAH02-05 - medical sciences > CAH02-05-02 - healthcare science (non-specific)
Divisions: Doctoral Research College > Doctoral Theses Collection
Faculty of Computing, Engineering and the Built Environment > College of Computing
Depositing User: Louise Muldowney
Date Deposited: 13 Jan 2025 11:27
Last Modified: 13 Jan 2025 11:27
URI: https://www.open-access.bcu.ac.uk/id/eprint/16068

Actions (login required)

View Item View Item

Research

In this section...