search

physimals / oxasl

Python based ASL pipeline based on oxford_asl
https://oxasl.readthedocs.io/
Apache License 2.0
5 stars 6 forks source link

Step 2 of 3: VB - Tissue Arterial - Initialise with step 1 Using fabber - Taking too much time #14

Open ampaschoal opened 3 years ago

ampaschoal commented 3 years ago

Hi all,

I was using the oxasl with FSL version 6.0.1 and it was running well. Last week I updated my FSL to 6.0.4 and when using oxasl (or oxford_asl), the Step 2 of BASIL analysis is taking too much time, so much that I cant finish (one hour for 1%). Here is the print of my try. I am using Ubuntu 18.04.

Did it happen with anybody else? What should I do to solve it?

I already tried to re-install my FSL as well as the oxasl.

Screenshot from 2020-09-11 15-35-41

ampaschoal commented 3 years ago

After installing back the 6.0.1 version of FSL, I could run again the oxasl.

mcraig-ibme commented 3 years ago

Sorry for the slow response, I have not been getting Github notifications on this repository. I am not aware of any problems with Fabber in FSL 6.0.4. Was the first step (which seems to have completed) also very slow, or did that finish in reasonable time?

ampaschoal commented 3 years ago

No, the first step was quick. The second step I couldn't finish. I waited for a few hours and it still was at the beginning. I have no clue why. But after installing back an older version of FSL it worked, so I didn't try again on version 6.0.4.

mcraig-ibme commented 3 years ago

My best guess is that there was a numerical instability in the model. This can cause Fabber to fill up its logfile with huge amounts of data which would certainly slow it down a lot. If you have a data sample that you are able to share then I might be able to investigate and reproduce it. The issue definitely looks to be in fabber itself which would fit with the change in behaviour on installing a different version of FSL.

robbwh commented 3 weeks ago

I similarly can only get the arterial inference to work (within a reasonable timeframe) when using FSL 6.0.1.

Using FSL 6.0.5.1 it runs so slowly that I doubt it would complete within 24 hours.

The reason for this is unclear to me. Using the fmrib example multi-pld data, oxasl/oxford_asl run will complete just fine regardless of version.

I am unable to share data currently, but the data come from a test subject acquired using the ADNI4 multi-pld pCASL scan card.

EDIT: if --bat=1.3 (which is default for pCASL, will not run). If bat is specified lower, like --bat=0.7 to --bat=1.25, it will run