A master thesis from Aalborg University
Validation of Synthesised T1-weighted Images from FLAIR
[Validering af syntetiseret T1-vægtet billeder fra FLAIR]
Author(s)
Term
4. term
Education
Publication year
2025
Submitted on
2025-06-02
Pages
46 pages
Abstract
Baggrund: Magnetisk Resonans billeddannelse (MR) anvendes hyppigt klinisk til vurdering af apopleksi, hvor Fluid Attenued Inversion Recovery (FLAIR) er en central sekvens til detektering og karakterisering af patologi. Dog er T1-vægtede sekvenser, som ofte kræves til avanceret neurobilledanalyse og forskningssoftwareværktøjer, ofte utilgængelige, hvilket udgør en betydelig begrænsning. SynthSR, et neuralt netværk der kan syntetisere manglende T1-vægtet billeder ud fra andre MR-sekvenser, tilbyder en potentiel løsning. Formålet med dette studie var at validere SynthSR’s evne til at syntetisere realistiske T1-vægtet billeder ud fra FLAIR, og at vurdere ligheden af de syntetiserede T1-vægtet billeder sammenlignet med referencedata. Metode: Data bestod af FLAIR og T1-vægtet MR-billeder fra 95 apopleksipatienter. FLAIR billederne fungerede som input til SynthSR for at syntetisere T1 billeder. Efter syntetisering, blev både de originale og syntetiserede T1 billeder registreret til MNI-152 skabelon og normaliseret før udregning af Mean Squared Error (MSE) og Structural Similarity Index (SSIM), samt subjektiv vurdering. Resultater: 87 FLAIR og T1-vægtet billede par blev inkluderet til syntetisering. Objektiv vurdering viste en gennemsnits MSE på 0.56±0.17 og en gennemsnits SSIM på 0.35±0.66 når de syntetiserede T1-vægtet billeder blev sammenlignet med de originale T1-vægtet billeder. Sammenlignet med MNI-152 skabelonen viste de originale T1-vægtet billeder en gennemsnits MSE på 1.12±0.17 og en gennemsnits SSIM på 0.32±0.06, hvor de syntetiserede T1-vægtet billeder viste en gennemsnits MSE på 0.99±0.04 og en gennemsnits SSIM på 0.32±0.01. Konklusion: SynthSR syntetiserede T1-vægtet billeder ud fra FLAIR der visuelt lignede de originale T1-vægtet billeder. Dog viste objektive målinger forskelle i både pixelintensitet og strukturel lighed, hvilket understreger behovet for yderligere forskning.
Background: Magnetic Resonance Imaging (MRI) is widely used clinically for stroke assessment, with Fluid-Attenuated Inversion Recovery (FLAIR) being a key sequence for detecting and characterising pathology. However, T1-weighted sequences, often required for advanced neuroimaging analysis and research software tools, are frequently unavailable, posing a limitation. SynthSR, a neural network capable of synthesising missing T1-weighted images from other MRI sequences, offers a potential solution. The aim of this study was to validate SynthSR’s capability to synthesise realistic T1-weighted images from FLAIR, and to assess the similarity of the synthesised T1-weighted images compared to the ground truth. Method: Data consisted of FLAIR and T1-weighted MRI images from 95 stroke patients. FLAIR images served as input for SynthSR to synthesise T1-weighted images. After synthesis, both the original and synthesised T1-weighted images were registered to the MNI-152 template and normalised before computing Mean Squared Error (MSE) and Structural Similarity Index (SSIM), alongside subjective validation. Results: 87 FLAIR and T1-weighted image pairs were included for synthesis. Objective assessment revealed a mean MSE of 0.56±0.17 and mean SSIM of 0.35±0.66 when synthesised T1-weighted images were compared against original T1-weighted images. Against the MNI-152 template, original T1-weighted images exhibited a mean MSE of 1.12±0.17 and mean SSIM of 0.32±0.06, whereas synthesised T1-weighted images showed a mean MSE of 0.99±0.04 and mean SSIM of 0.32±0.01. Conclusion: SynthSR synthesised T1-weighted images from FLAIR that were visually similar to the original T1-weighted images. However, objective measurements revealed differences in both pixel intensity and structural similarity, emphasising the need for further research.
Documents
Colophon: This page is part of the AAU Student Projects portal, which is run by Aalborg University. Here, you can find and download publicly available bachelor's theses and master's projects from across the university dating from 2008 onwards. Student projects from before 2008 are available in printed form at Aalborg University Library.
If you have any questions about AAU Student Projects or the research registration, dissemination and analysis at Aalborg University, please feel free to contact the VBN team. You can also find more information in the AAU Student Projects FAQs.