Validation of Synthesised T1-weighted Images from FLAIR
Translated title
Validering af syntetiseret T1-vægtet billeder fra FLAIR
Author
Rasmussen, Maria Dalmose
Term
4. term
Education
Publication year
2025
Submitted on
2025-06-02
Pages
46
Abstract
Baggrund: MR-skanning (MRI) bruges ofte til at vurdere slagtilfælde. FLAIR er en vigtig skanningssekvens til at opdage forandringer i hjernen, mens T1-vægtede billeder ofte kræves for avancerede analyser og forskningsværktøjer. I praksis mangler T1-billeder dog tit. SynthSR er et neuralt netværk, der kan genskabe manglende T1-billeder ud fra andre MR-sekvenser. Formålet med dette studie var at undersøge, om SynthSR kan lave realistiske T1-billeder ud fra FLAIR, og hvor tæt de kommer på de oprindelige T1-billeder. Metode: Data stammede fra 95 patienter med slagtilfælde. FLAIR-billeder blev brugt som input til SynthSR for at syntetisere T1-billeder. Både de oprindelige og de syntetiske T1-billeder blev derefter tilpasset en standard-hjerneskabelon (MNI-152) og normaliseret. Vi sammenlignede dem ved hjælp af mål for pixel-forskel (MSE, middelkvadratfejl) og strukturel lighed (SSIM) og lavede også en visuel vurdering. I alt indgik 87 ud af 95 datasæt i syntesen. Resultater: De syntetiske T1-billeder så visuelt ud til at ligne de oprindelige T1-billeder. De objektive mål viste dog forskelle i både pixelintensitet og strukturel lighed, når de syntetiske billeder blev sammenlignet med de oprindelige. Sammenlignet med MNI-152-skabelonen havde de oprindelige og de syntetiske T1-billeder lignende strukturel lighed, mens de syntetiske viste lidt lavere pixelafvigelse. Konklusion: SynthSR kan lave T1-billeder fra FLAIR, som ser realistiske ud, men de adskiller sig målbart fra de oprindelige T1-billeder. Der er behov for yderligere forskning, før metoden kan erstatte manglende T1-billeder i avancerede analyser.
Background: MRI is widely used to assess stroke. FLAIR is a key sequence for detecting brain changes, while T1-weighted images are often required for advanced analyses and research tools. In practice, T1 images are frequently missing. SynthSR is a neural network that can reconstruct missing T1 images from other MRI sequences. This study aimed to test whether SynthSR can create realistic T1 images from FLAIR and how closely they match the original T1 images. Method: Data came from 95 stroke patients. FLAIR images were used as input to SynthSR to synthesise T1-weighted images. Both original and synthesised T1 images were aligned to a standard brain template (MNI-152) and normalised. We compared them using measures of pixel-level difference (MSE, mean squared error) and structural similarity (SSIM), and we also performed a visual review. In total, 87 of the 95 datasets were included in the synthesis. Results: The synthesised T1 images appeared visually similar to the original T1 images. However, objective metrics showed differences in both pixel intensity and structural similarity when the synthesised images were compared with the originals. Against the MNI-152 template, original and synthesised T1 images showed similar structural similarity, while the synthesised images had slightly lower pixel deviation. Conclusion: SynthSR can generate realistic-looking T1 images from FLAIR, but measurable differences remain compared with real T1 images. Further research is needed before this approach can replace missing T1 images in advanced analyses.
[This summary has been rewritten with the help of AI based on the project's original abstract]
Documents