Intriguing! Okay, I don't see anything obviously wrong (to my meager eyes) in the prestats; motion levels are reasonable, and we've run MCFLIRT motion correction. So, here's a full .feat directory from one of the "first step" analyses (i.e., pre-stats). I've included the raw functional, fieldmap, and structural files too.
In case the workflow is unclear, I used this first step analysis to get a filtered_func_data that's had prestats/MELODIC. I looked through the MELODIC results to find noise components, and removed them with fsl_regfilt to create denoised_data. That denoised_data file was the input to the second step of the first-level analysis (stats & post-stats).
New files at: https://www.dropbox.com/s/el79bfvevdnbe1p/prestatsPhilip.zip
Thanks,
-Benjamin Philip
On Tue, 25 Jun 2013 21:54:28 +0000, Mark Jenkinson <[log in to unmask]> wrote:
>Hi,
>
>I've had a look at the images that you sent and I am more convinced than ever that this is not a registration problem. If you look at the activations in the first-level results (e.g. run1_zstat2) then you see just as much activation outside the brain as you do in the second-level results. Given that the first-level results are shown on the functional images *prior* to any registration, then that means that it is not the registration that is causing these activations outside of the brain.
>
>So that means that the cause of these activations still needs to be found. It might be due to motion artefacts (and I couldn't tell if you were running motion correction or not from your FEAT design files) or very strong smoothing or maybe actual venous BOLD changes. It would therefore be helpful to know more about what pre-processing steps you are using for this data, and whether there is a lot of motion, or anything unusual, in this dataset.
>
>All the best,
> Mark
>
>
>
>On 25 Jun 2013, at 17:49, Benjamin Philip <[log in to unmask]> wrote:
>
>> I agree that it's a confusing situation. If I understood the problem I could do more of the troubleshooting myself! My interpretation/guess has been that the 1st-level registrations are doing something wrong that I can't discern by eye. But if the 2nd-level analysis included a reg step, that would be a much more sensible location for the problem, wouldn't it?
>>
>> I've uploaded reg, reg_standard, and zstat images for all 4 first-level runs; and the matching zstat from the second-level analysis. You can find them at: https://www.dropbox.com/s/nmqm7rulbyxaa65/regTestPhilip4.zip
>>
>> Thanks,
>>
>> -Benjamin Philip
>>
>> On Mon, 24 Jun 2013 21:05:16 +0000, Mark Jenkinson <[log in to unmask]> wrote:
>>
>>> Dear Benjamin,
>>>
>>> I'm still confused about this. You say that the first-level registrations look fine but the second-level (still within-subject) are not. I think I need to see the first-level and second-level statistics for one particular contrast in one example subject. So can you upload the raw zstat files (not the rendered version) from one selected contrast from both the first and second levels, together with the registration folders. Note that I'd like to see the first-level results from each session that feeds into the second-level (still one subject and one contrast).
>>>
>>> Hopefully I can see what is happening to the first-level stats when they are transformed and combined in the second-level.
>>>
>>> All the best,
>>> Mark
>>>
>>>
>>> On 18 Jun 2013, at 19:59, Benjamin Philip <[log in to unmask]> wrote:
>>>
>>>> Mark-
>>>> In my pipeline, I did my first-level analysis in two steps: a pre-stats analysis including unwarping and denoising, followed by a second step that included stats and standard-brain registration. The included .fsf file (motor1.fsf) is the second step only - you may note it's set for "Stats + Post-stats only", thus it shouldn't apply the fieldmap again - which would indeed, as you note, cause problems. It's a good idea, but I believe I'm already avoiding this pitfall.
>>>>
>>>> But in the context of fieldmap-only-applied-once: Even if example_func2standard looks okay, a second-level analysis based on these data still produce terrible registration results. Since there's no second-level reg directory, I'm not sure what information I can provide to help you identify the problem. So, for the moment, here's something that demonstrates the problem: I've uploaded a second-level rendered_thresh_zstat file to demonstrate the immense non-brain activity. Since it was just a single file, I put it on the upload link you provided previously (https://oxfile.ox.ac.uk/oxfile/work/extBox?id=68312615463381F4C)
>>>>
>>>> Apologies on the deleting of threads - I hadn't figured out how to include a quote on the Jiscmail web interface. Got it now, obviously.
>>>>
>>>> Thank you for your continuing help - I really appreciate it!
>>>>
>>>> -Benjamin Philip
>>>>
>>>> On Tue, 18 Jun 2013 15:16:18 +0000, Mark Jenkinson <[log in to unmask]> wrote:
>>>>
>>>>> Dear Benjamin,
>>>>>
>>>>> I've tried several different options with your data (fieldmaps applied in +x/-x/+y/-y and no fieldmaps). In the end the result with no fieldmaps is clearly the best. I notice that your input data is called "denoised_data" - so have you run this through FEAT/MELODIC or some other denoising process that might already have corrected for the fieldmap distortions? That would explain why it works worse with fieldmaps now, as that would be applying the correction twice. Alternatively, maybe there is something wrong in the scaling of your fieldmaps, or the echo time difference, such as not taking into account parallel accelerations.
>>>>>
>>>>> Try it with no fieldmap correction and see the results yourself. There is still some intensity outside of the outlines of the standard brain in the example_func2standard, but this is to be expected with the peripheral CSF that you see in your functional images.
>>>>>
>>>>> All the best,
>>>>> Mark
>>>>>
>>>>> P.S. Can I please ask you not to erase the previous thread from your emails in future, as it makes it harder for us to trace back what has been said and tried before.
>>>>>
>>>>>
>>>>> On 17 Jun 2013, at 07:03, Benjamin Philip <[log in to unmask]> wrote:
>>>>>
>>>>>> I believe I am still waiting for a reply to this question? As noted in my post on June 6th, I've applied both of the fixes that Mr. Jenkinson suggested, but I'm still getting registrations with the same substantial non-brain activation (especially at the second level). I'm looking forwards for suggestions to how I might troubleshoot this problem. If any files or information are needed beyond those offered in the June 6 post, please let me know.
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>> Benjamin Philip
|