Signal-to-Noise in MRI

[Pages:21]Signal-to-Noise in MRI

? What is SNR? ? Noise statistics & Measurement ? Multichannel noise

1

Section B4

B.Hargreaves - RAD 229

Basic Noise Statistics

=0

(mean)

(See conceptB4_2.m)

2

Section B4

B.Hargreaves - RAD 229

FFT of Gaussian Noise

? Note sqrt(N) scaling preserves noise energy

3

Section B4

B.Hargreaves - RAD 229

Basic SNR Measurement (1 coil)

? Measure mean in signal area ROI ? Measure std-deviation in magnitude background ROI ? Correct for Rayleigh distribution in background

meanRayleigh = 1.26 Rayleigh = 0.65

4

gaussian = meanRayleigh / sqrt(/2) = 1.008 gaussian = Rayleigh / sqrt(2-/2) = 0.997

Section B4

B.Hargreaves - RAD 229

Difference Method SNR

? In theory, N measurements should give you a population, and at each pixel you get a (roughly gaussian) distribution

? With 2 measurements you can still estimate mean and standard deviation (Reeder et al)

Mag-Diff = 1.394 gaussian = Mag-Diff/ sqrt(2)

Sum

Difference of Magnitude Images

5

Section B4

B.Hargreaves - RAD 229

Multiple Coils

? Multiple images, ideally with uncorrelated noise ? Combine with RMS or sensitivity-based methods

? Coils

6

Section B4

B.Hargreaves - RAD 229

Multi-Coil Combinations

? Si = signal from coil i, Ci = sensitivity of coil i

?

Root-mean-square

(RMS)

SRM S

=

v u u tNX coils

Si2

i1

? SENSE (Each pixel, for any reduction factor R)

S = mC

SSENSE = CH 1C 1 CH 1S

X N coils

SSENSE =

iSi

i=1

7

Section B4

B.Hargreaves - RAD 229

Multiple Coils

Single-Channel

RMS

SENSE

(uniform noise) (Signal shading) (Noise Varies)

8

Section B4

B.Hargreaves - RAD 229

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download