Berlin 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz
AKPIK 1: Reservoir Computing & Neural Networks
AKPIK 1.4: Vortrag
Dienstag, 19. März 2024, 10:15–10:30, MAR 0.002
Image reconstruction with diffusion models for accelerated magnetic resonance imaging — •Christine Müller1, Vanya Saksena2, Florian Knoll2, and Bernhard Kainz1 — 1Image Data Exploration and Analysis Lab, Friedrich-Alexander-Universität Erlangen-Nürnberg — 2Computational Imaging Lab, Friedrich-Alexander-Universität Erlangen-Nürnberg
Magnetic resonance imaging (MRI) is a widely used non-invasive imaging technique in medical practice. However, conventional MRI acquisitions are often time-consuming, limiting their applicability in clinical settings. Accelerated MRI techniques reduce scan time but come at the expense of image quality. Diffusion models are a recent promising class of generative models for image reconstruction in MRI [1,2]. These models learn to generate high-quality images from undersampled MRI data by gradually denoising a noisy image.
In this work, we evaluate the performance of a score-based diffusion model that was trained on brain MRI data from the publicly available fastMRI dataset [3]. It employs a predictor-corrector sampling step and combines parallel imaging and compressed sensing techniques. A learned score function is utilized as a prior to guide the reconstruction process, enabling the generation of realistic content, especially at high acceleration rates. Diffusion models offer the potential to improve the quality of reconstructed images and reduce scan time, making MRI more accessible and efficient.
[1] DOI: 10.1016/j.media.2022.102479, [2] DOI: 10.1016/ j.media.2023.102846, [3] DOI: 10.1148/RYAI.2020190007
Keywords: diffusion models; magnetic resonance imaging; image reconstruction; inverse problems