search

firasm / CEST

Analysis of studies
2 stars 0 forks source link

Effect of Slice Thickness #28

Closed srveale closed 9 years ago

srveale commented 9 years ago

unknown

unknown-1

DrSAR commented 9 years ago

looks like exquisite SNR. Also, is it fair to say that at slice thickness 4 (mm?) and above you have essentially reached the same SNR?

firasm commented 9 years ago

Scott you're acquiring all the way to 0.25 mm right?

Can you see if we can go any lower ?

On Fri, Mar 27, 2015 at 8:10 PM, srveale notifications@github.com wrote:

effectslicethickness

Reply to this email directly or view it on GitHub: https://github.com/firasm/analysis/issues/28

srveale commented 9 years ago

Yep, just got caught by range(11,16) not including 16.

On Fri, Mar 27, 2015 at 8:18 PM, firasm notifications@github.com wrote:

Scott you're acquiring all the way to 0.25 mm right?

Can you see if we can go any lower ?

On Fri, Mar 27, 2015 at 8:10 PM, srveale notifications@github.com wrote:

![effectslicethickness]( https://cloud.githubusercontent.com/assets/7737922/6879645/26451e94-d4bd-11e4-95bc-99d927b9895f.png

)

Reply to this email directly or view it on GitHub: https://github.com/firasm/analysis/issues/28

— Reply to this email directly or view it on GitHub https://github.com/firasm/analysis/issues/28#issuecomment-87153587.

firasm commented 9 years ago

Wow that's a huge difference between 1 and 0.25! 

Sorry to do this, but can you do 0.5 and something between 0.5 and 0.25? I want to see if the change is gradual or not. 

You can skip the power level experiment if you like in lieu of this.

On Fri, Mar 27, 2015 at 8:31 PM, srveale notifications@github.com wrote:

effectslicethickness

Reply to this email directly or view it on GitHub: https://github.com/firasm/analysis/issues/28#issuecomment-87154028

srveale commented 9 years ago

Yep I can try that. The power experiment is a third done. Should I scrap it and go for that instead?

Also, riddle me this: SNR drops with a slice thickness of 12....

snrresponsetoslicethickness

firasm commented 9 years ago

Up to you, do as much as you like. 

There must be something wrong witht the regions we're providing to the SNR function

On Fri, Mar 27, 2015 at 8:36 PM, srveale notifications@github.com wrote:

Yep I can try that. The power experiment is a third done. Should I scrap it and go for that instead? Also, riddle me this: SNR drops with a slice thickness of 12....

snrresponsetoslicethickness

Reply to this email directly or view it on GitHub: https://github.com/firasm/analysis/issues/28#issuecomment-87154754

DrSAR commented 9 years ago

I would expect near linear dependence of SNR on slice thickness. That, of course depends a bit on how SNR is measured. One way I could think of is SNR per voxel where the mean of signal in a offset-frequency region is defined as the signal and the noise from its standard deviation.

How do you do this?

srveale commented 9 years ago

I've been doing

SNR = average(signal)/stdev(noise)

where signal is the data from an roi and noise is the data from a region outside the phantom holder. I tried a few regions inside different phantoms and the SNR vs SliceWidth curve was similar for most of them.

On Fri, Mar 27, 2015 at 9:29 PM, DrSAR notifications@github.com wrote:

I would expect near linear dependence of SNR on slice thickness. That, of course depends a bit on how SNR is measured. One way I could think of is SNR per voxel where the mean of signal in a offset-frequency region is defined as the signal and the noise from its standard deviation.

How do you do this?

— Reply to this email directly or view it on GitHub https://github.com/firasm/analysis/issues/28#issuecomment-87159533.

DrSAR commented 9 years ago

That's a reasonable thing to do. It suffers from potentially large noise in the 'signal-empty' region if there is some sort of imaging artefact at low signal levels (e.g. a faint ghost). If you need to know SNR precisely and you've got time on your hands, you would re-acquire the same image over and over and then you look at mean/std for any pixel.

firasm commented 9 years ago

Long awaited results on what happens to the noise in the CEST spectrum as you reduce the SNR.

The plot is posted above in the issue (since that's where it belonged in the first place) but I've included it here for the benefit of people just checking this by email:

unknown-1

From here it is clear that the noise in the CEST spectrum increases considerably below a slice thickness of 1mm. The in-plane resolution was fixed at 0.78mm x 0.78mm.

Just to remind you, the mouse brain experiment last week that did not work had an in-plane resolution was 0.47mm x 0.47mm x 1mm.

DrSAR commented 9 years ago

one of those often-lamented truths: signal proportional to the number of protons