Quantification of variability sources and evaluation of myocarditis extent from MRI LGE imagesKralj, Lana (Avtor) Jeraj, Robert (Mentor) Kirn, Borut (Komentor) myocarditisquantitative MRI imagingMRI LGE image analysisvariabilityquantification methodsregion of interest (ROI)Introduction Clinical experts determine the extent and distribution of myocardial damage in myocarditis by quantitatively analyzing magnetic resonance late gadolinium enhancement (MRI LGE) images. The variability of results originates from two quantitative imaging steps: image acquisition and image analysis. The latter is studied in this master's thesis to rank all analysis steps with respect to their contribution to overall variability. Methods Whilst analyzing MRI LGE images of 35 myocarditis patients with the clinical experts, I determined three analysis steps, all three of them being selection of reference regions of interest (ROI): delineation of myocardium (ROI_myoc), reference pathological or healthy tissue (ROI_ref), and excluded tissue (ROI_excl). To visualize ROI’s and calculate LGE extent, I developed a quantitative MRI LGE image analysis tool. Using developed tool, four experts analyzed five images in three sessions, whereby LGE extent was quantified with two methods: 3σ, which uses signal intensity threshold of three standard deviations above ROI_ref defined in normal myocardium, and FWHM, which uses threshold of 50% of the maximum signal intensity within the scar-defined ROI_ref for pathology determination. Results and discussion I observed artefacts and image resolution impacted ROI_myoc and ROI_excl delineation, and that image contrast was crucial when selecting ROI_ref. These factors, in combination with subjective contour delineation during the analysis process, caused variations in ROI regions selection. The variability of ROI_myoc and ROI_excl was significantly smaller when using the FWHM method. Furthermore, ROI_ref variability was significantly greater than ROI_myoc variability when using the 3σ method. It was demonstrated that overall variability may exceed clinically acceptable 5% limit, with the risk being greater when using the 3σ method, which could lead to misevaluation of the adverse effects risk. Relative shares of ROI variabilities showed that ROI_ref contributes on average 1.8-times more to overall variability than ROI_myoc in the 3σ method. Thereby, clinical experts should be especially careful when selecting ROI_ref.20212021-09-12 08:15:06Magistrsko delo/naloga130268sl