Learning a Sampling Pattern for MRI Ferdia Sherry Supervised by - - PowerPoint PPT Presentation

Learning a Sampling Pattern for MRI

Ferdia Sherry

Supervised by Carola-Bibiane Sch¨

• nlieb and Matthias J. Ehrhardt

May 25, 2017

Ferdia Sherry Learning a sampling pattern for MRI May 25, 2017 1 / 6

Magnetic Resonance Imaging

(a) An MRI machine in

• peration

(b) An image of a knee acquired by MRI

We model the measurements y taken by the MRI machine as y = SFu + ε and reconstruct the image from measurements y by minimising a TV-regularised least squares functional: u∗ = argmin

u′0

ETV(u′, y, S, α).

Ferdia Sherry Learning a sampling pattern for MRI May 25, 2017 2 / 6

The effect of the sampling pattern

Ferdia Sherry Learning a sampling pattern for MRI May 25, 2017 3 / 6

Learning a sampling pattern

Given a training set of ground truth images {u1, . . . , un} and corresponding measurements {y1, . . . , yn} we want to determine a suitable sampling pattern S for similar images. To do this we study the bilevel problem min

S,α

1 n

n

• i=1

ℓ(u∗

i , ui) + penalty(S, α)

where u∗

i = argmin u0

ETV(u, yi, S, α) with ℓ a loss function measuring the reconstruction error.

Ferdia Sherry Learning a sampling pattern for MRI May 25, 2017 4 / 6

A learned sampling pattern and reconstruction

Ferdia Sherry Learning a sampling pattern for MRI May 25, 2017 5 / 6

Thank you for listening!

Ferdia Sherry Learning a sampling pattern for MRI May 25, 2017 6 / 6