Methods for assisting the automation of Dynamic Susceptibility Contrast Magnetic Resonance Imaging Analysis

Sobhan, Rashed (2020) Methods for assisting the automation of Dynamic Susceptibility Contrast Magnetic Resonance Imaging Analysis. Doctoral thesis, University of East Anglia.

[thumbnail of SobhanRPhD.pdf]
Download (3MB) | Preview


Dynamic susceptibility-contrast magnetic resonance imaging (DSC-MRI) is widely used for cerebral perfusion measurement, but dependence on operator input leads to a time-consuming, subjective, and poorly-reproducible analysis. Although automation can overcome these limitations, investigations are required to further simplify and accelerate the analysis. This research focuses on automating arterial voxel (AV) and brain tissue segmentation, and model-dependent deconvolution steps of DSC-MRI analysis.

Several features were extracted from DSC-MRI data; their AV- and tissue voxel- discriminatory powers were evaluated by the area-under-the-receiver-operating-characteristic-curve (AUCROC). Thresholds for discarding non-arterial voxels were identified using ROC cut-offs.

The applicability of DSC-MRI time-series data for brain segmentation was explored. Two segmentation approaches that clustered the dimensionality-reduced raw data were compared with two raw−data-based approaches, and an approach using principal component analysis (PCA) for dimension-reduction. Computation time and Dice coefficients (DCs) were compared.

For model-dependent deconvolution, four parametric transit time distribution (TTD) models were compared in terms of goodness- and stability-of-fit, consistency of perfusion estimates, and computation time.

Four criteria were effective in distinguishing AVs, forming the basis of a framework that can determine optimal thresholds for effective criteria to discard tissue voxels with high sensitivity and specificity.

Compared to raw−data-based approaches, one of the proposed segmentation approaches identified GM with higher (>0.7, p<0.005), and WM with similar DC. The approach outperformed the PCA-based approach for all tissue regions (p<0.005), and clustered similar regions faster than other approaches (p<0.005).

For model-dependent deconvolution, all TTD models gave similar perfusion estimates and goodness-of-fit. The gamma distribution was most suitable for perfusion analysis, showing significantly higher fit stability and lower computation time.

The proposed methods were able to simplify and accelerate automatic DSC-MRI analysis while maintaining performance. They will particularly help clinicians in rapid diagnosis and characterisation of tumour or stroke lesions, and subsequent treatment planning and monitoring.

Item Type: Thesis (Doctoral)
Faculty \ School: Faculty of Medicine and Health Sciences > Norwich Medical School
Depositing User: Chris White
Date Deposited: 14 Jul 2021 14:47
Last Modified: 14 Jul 2021 14:47


Downloads per month over past year

Actions (login required)

View Item View Item