CRLB for Osprey modelling

MRS Software Osprey
1

Hi,
I’m wondering if there is any way to get the Cramer-Rao lower bounds for a fit in Osprey. Alternatively, what type of error estimations in the fit parameters are available in Osprey?.
Thanks,
/Greger

2

Hi Greger,

We’ve been working hard on getting CRLBs returned correctly, but there are still glitches when we calculate the Jacobian matrix… it’s something that keeps haunting us. We’re aware that’s quite the shortcoming for now, but hope to have it fixed at some point. You’ll read it here :slight_smile:
Best,
Georg

3

Hi @admin !
Any update about the CRLB values?
Best,
Frederic

4

Not at this point. We’re planning some larger changes to the fit module this year, and we’re certainly eager to include CRLBs in those changes, too.

5

Very excited to see what you all have planned!

1 Like
6

Hello,

I was looking for info on CRLB and came across this thread. Is there any news regarding this topic?

BW,
Andreia

7

Not yet, and I suspect that @admin will support this with articles (below) explaining that CRLB is not an adequate approximation to standard deviation in MRS. :wink:

Best,
Frederic

Roland Kreis.The trouble with quality filtering based on relative Cramér-Rao lower bounds.2016
Karl Landheer, Christoph Juchem.Are Cramér-Rao lower bounds an accurate estimate for standard deviations in in vivo magnetic resonance spectroscopy?2021

8

Hi Frederic,

Thanks for your reply, it makes sense indeed!

@admins Are there any plans to include some type of error estimation for metabolite quantification? Or how do you recommend that one report on Osprey’s fitting quality?

Thanks,
Andreia

9

@admin Can I please check how you would recommend reporting quality in Osprey? I understand that CRLB% is not appropriate, but should absolute CRLB still be reported (and if so will Osprey provide it)? Is reporting Cr SNR, Cr FWHM, water FWHM, and freqShift adequate? Many thanks.

1 Like
10

Hi, I also wanted to follow up on this topic and I am interested in reporting data quality. Is there also an option for linebroading?
Thank you. Best,
Stephanie

11

  1. The current version of Osprey does not provide CRLB, unfortunately. We are testing a new release currently which will come with a completely new modeling module including 2D fitting and, finally, properly validated CRLB. @jbarnacle Yes, I would report SNR and FWHM; Osprey also gives you a relative fit quality metric (calculated by comparing the maximum residual to the standard deviation of the noise).

  2. There is no option for linebroadening. That is on purpose. Linebroadening (and, btw, zero-filling beyond a factor of 2) destroys the independence of data points and therefore violates assumptions made during linear-combination modeling (more precisely, the least-squares solution is identical to the maximum-likelihood estimate only if the noise is independent and identically distributed - to be fair, I’m not aware how bad it is to violate these assumptions and how far off the solutions would be - someone should write a paper about this).

3 Likes
12

@admin can you let us know when the new release will be ready? Thank you so much!

1 Like
13

Hi! Sorry, where can we find the relative fit quality metric?

14

Never mind! I read on another post that it is stored under the relResA column in QM_processed_spectra. Posting it in here in case anyone else has the same question.

15

@admin, talking about relative fit quality metric, I wanted to double confirm my interpretation of the relative residual is correct. If my average relAmpl value in the Osprey container is 3, that means my relative amplitude of the residual is 3 times of the standard deviation of the noise, right?

Thanks.
Steve