FBER scores of -1

[ostine]

I’m having subjects get a fiber score of -1. Not sure what the cause is. Found many with this error: estimated signal variation in background is too small. Not all of those with the error got a fber of -1. Thoughts? What could cause this (score of -1)? And is it related to the error?

[jblujus]

Hello,

I am also interested in this topic. Has anyone determined the cause of a FBER score of -1 (MRIQC of T1w data)?

Best,
Jenna

[rwblair]

Here’s the code from mriqc for calculating FBER. Looks like if the median of the square of airmask is less than .001 it will return -1:

    fg_mu = np.median(np.abs(img[headmask > 0]) ** 2)

    airmask = np.ones_like(headmask, dtype=np.uint8)
    airmask[headmask > 0] = 0
    if rotmask is not None:
        airmask[rotmask > 0] = 0
    bg_mu = np.median(np.abs(img[airmask == 1]) ** 2)
    if bg_mu < 1.0e-3:
        return -1.0
    return round(float(fg_mu / bg_mu), decimals)

Link to github

[Bartosz_Kossowski]

Hi,
I’m experiencing the same issue with our new short T1w sequence. I’ve tried modifying the code above, for example, setting bg_mu < 1.0e-9 but with no success. I do observe low but reasonable background noise, as shown in the example image and group plot. I also tried an alternative method, RadQA, which provides a very meaningful fBER that aligns well with our visual quality scores.