casperdcl / brainweb

BrainWeb: Multimodal models of 20 normal brains

^{_{Open source lines written (click for more info):}}

TL;DR blog	Projects	Teaching	Talks	Academia	Sponsors & perks

clients & employers include: @sony | @worldbank | @MetOffice | @UCL | @stfc | @iterative | @HawkEyeInnovations | @openworm | UKRI | Imperial | LSE | KCL

brainweb's Issues

brainweb "index" values do not match

the imported label image has different values than expected, leading to aroundFat, dura and marrow being empty

>>> import sirf.STIR
>>> im=sirf.STIR.ImageData('brainweb_labels.nii')
>>> arr=im.as_array()
>>> numpy.unique(arr)
array([  0.,  16.,  32.,  48.,  64.,  80.,  96., 112., 128., 145., 161., 177.], dtype=float32)
>>> import brainweb
>>> brainweb.Act.aroundFat
144
>>> brainweb.Act.dura
160
>>> brainweb.Act.marrow
176

It seems safer to use the 8bit valued images, in which case we wouldn't have this problem

allow specifying different resolution

toPetMmR has an an outres parameter

brainweb/brainweb/utils.py

Lines 398 to 399 in 202eb40

	def toPetMmr(im, pad=True, dtype=np.float32, outres="mMR", modes=None,
	PetClass=FDG):

It would be nice to be able to pass this to the higher level get_mmr and get_mmr_fromfile as well, such that we can get images at different voxel-size out.

It would be even better to be able to specify arbitrary voxel-size/shape, but that seems to only be possible by modifying the Res/Shape classes, which seems to require major surgery, or possibly doing a setattr(Res, mMR, ...) and same for Shape

get_mmr_fromfile cache - function returns same results with different noise/sigma args

The cache file in get_mmr_fromfile() is generated based only on a subset of the input parameters (specifically, not the noise level/sigma). This can also be surprising for the user that changing the noise sigmas results in the same image.

Issues with resolution other than mMR

G'day Casper,

Loving this tool, thanks.

There are a few issues with using resolutions other than mMR.

The Shape class should cast its calculations to type np.int (and floor?)
Otherwise I run into type issues here: https://github.com/casperdcl/brainweb/blob/master/brainweb/utils.py#L301

Somewhat related but not really: the noise() function assumes sigma is given in voxel units - is this preferred? Potentially confusing for the user, we had to change sigma with resolution.

get_mMR_from_file ignore parameters?

I believe these are wrong

brainweb/brainweb/utils.py

Lines 259 to 260 in 202eb40

	petNoise=1.0, t1Noise=0.75, t2Noise=0.75,
	petSigma=1.0, t1Sigma=1.0, t2Sigma=1.0,

as they overwrite any arguments specified to the function, but should be

petNoise=petNoise, t1Noise=t1Noise, t2Noise=t2Noise, 
 petSigma=petSigma, t1Sigma=t1Sigma, t2Sigma=t2Sigma, 

Problem with PET image plot in brainweb

Hi Casper,

When I'm trying to run brainweb, as PET output I am getting a blank plot.

Please see the screenshot as attached.

Could you please assist to fix it. I tried running the code many times.

`volshow` broken in `matplotlib>=3.3.2`

using matplotlib==3.3.2, the brainweb.volshow slider doesn't work. Changing the slider value doesn't change slices, but instead creates a new blank figure.

Everything works as expected in matplotlib==3.3.1.

toPetMmr ignores modes argument

it will always compute [PET, Mu, T1, T2]

extract labelfields

is there an easy way to extract labelfields for the different VOIs (white matter, grey matter, skin, etc.)? End goal is to create a 4D .nii with a 3D image per VOI. Background would also be nice. The values can be binary or probabilistic, we don't mind (as long as sum =1).

Our initial thought was to use petNoise=0, petSigma=0 and then do value-based labelfields, but I'm not sure what happens when you have a mixture of two VOIs in the same voxel.