tag:blogger.com,1999:blog-5737874959005852552.comments2024-03-12T11:19:30.567-05:00MVPA MeanderingsJo Etzelhttp://www.blogger.com/profile/04277620767760987432noreply@blogger.comBlogger249125tag:blogger.com,1999:blog-5737874959005852552.post-8922875259156116802023-09-15T06:35:52.599-05:002023-09-15T06:35:52.599-05:00Very useful thank you!Very useful thank you!Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-62618521559818048992023-08-25T15:38:02.603-05:002023-08-25T15:38:02.603-05:00Glad it's useful! The files are in https://osf...Glad it's useful! The files are in https://osf.io/w7zkc/. Go to the "Files" part of the page, and expand "OSF Storage (United States)". If you click on "knitr tutorials" it should enable a "Download a zip" button to download all the files at once, or you can click and download the individual files.Jo Etzelhttps://www.blogger.com/profile/04277620767760987432noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-56829973143050731462023-08-23T09:35:11.376-05:002023-08-23T09:35:11.376-05:00Dear Jo,
Thank you so much for all your content,...Dear Jo, <br /><br />Thank you so much for all your content, it is helping me enormously. I have a question, in which repository do you find the standard fsaverage5 .surf.gii, pial, etc., files?<br /><br />Thank you,<br />LuizaAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-53704512749692172012023-08-16T16:53:03.352-05:002023-08-16T16:53:03.352-05:00Which code exactly are you looking for? Nikolaus K...Which code exactly are you looking for? Nikolaus Kriegeskorte should have access to all the current LD-t; I haven't been following its development closely, and it wouldn't surprise me if they've gone further with the methods since I've last looked at it.Jo Etzelhttps://www.blogger.com/profile/04277620767760987432noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-36188627652501692962023-08-08T05:06:37.906-05:002023-08-08T05:06:37.906-05:00Dear Dr. Etzel,
If possible, can you please share...Dear Dr. Etzel,<br /><br />If possible, can you please share the code that utilizes the steps explained in this article? (Sorry for double post, I forgot to check "notify me")Tamer Gezicihttps://www.blogger.com/profile/11941973234066508233noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-55462035863760110682023-08-08T05:05:38.821-05:002023-08-08T05:05:38.821-05:00This comment has been removed by the author.Tamer Gezicihttps://www.blogger.com/profile/11941973234066508233noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-28559870894970156652022-07-13T21:06:31.478-05:002022-07-13T21:06:31.478-05:00Ah, this (moving between runs) is quite possible. ...Ah, this (moving between runs) is quite possible. I will ask whether we re-shim at the start of each run; I don't know. We do regularly advise people that if they must adjust their position to do so between, rather than during runs. We've never seen anything this dramatic before, but have occasionally seen a "dim" run in an otherwise normal session; I will investigate further.Jo Etzelhttps://www.blogger.com/profile/04277620767760987432noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-69170558479578543022022-07-13T18:52:14.997-05:002022-07-13T18:52:14.997-05:00Do you re-shim prior to each run? I'm thinking...Do you re-shim prior to each run? I'm thinking the motion may actually be between runs, possibly during the pre-scan adjustments (on-res adjustment and shim, if performed). According to your side by side comparisons, there's quite a substantial difference in the absolute brain location between runs #2 and #4. So even if the top half of the coil was performing normally, if the subject moved during a re-shim and the axis of rotation was near the occipital region, we might expect the largest problems in frontal regions. praciCal fMRIhttps://www.blogger.com/profile/08592275879035533857noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-77125571343128567332022-07-13T12:11:37.137-05:002022-07-13T12:11:37.137-05:00We're not using head cases for this study, but...We're not using head cases for this study, but have been seeing more movement than we'd like, and so trying to be more careful with packing; entirely possible this person was quite firmly packed. Reviewing their realignment parameters, they have pretty substantial apparent motion but not that much actual. <br />Jo Etzelhttps://www.blogger.com/profile/04277620767760987432noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-49854094394284309892022-07-13T11:42:23.440-05:002022-07-13T11:42:23.440-05:00Actually, there may be a simpler explanation: moti...Actually, there may be a simpler explanation: motion. There are two things that suggest motion. First, note that the problem image has highest SNR in the occipital region, the opposite of the normal images where frontal lobe is hottest. Second, there is an increased distortion in the frontal region in the problem image. Note the extra stretch at the most anterior portion of the brain, and a small signal "divot" that is not apparent in the normal images. Now we consider why there might be scalp ghosts over the brain. These could result from residual aliasing because of a poor Rx field. All these observations are consistent with the top half of the RF coil being displaced and connected improperly for the duration of the run. I've seen instabilities result from printed head cases (see https://practicalfmri.blogspot.com/2021/02/restraining-32-channel-coil.html) and analogous problems could result from a head very firmly packed in with foam. Given that motion is the usual suspect, I would investigate motion as the most likely cause here, too!praciCal fMRIhttps://www.blogger.com/profile/08592275879035533857noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-59441627169456196232022-07-13T02:05:27.963-05:002022-07-13T02:05:27.963-05:00Thats a cool artifact, thanks for sharing. I have ...Thats a cool artifact, thanks for sharing. I have never seen such an artifact in my own data. However, I think that it could be explained if something went wrong during the frequency adjust, which is usually done at the beginning of each run. This could result in an insufficient fat-saturation pulse and weaker excitation pulse?- Renzo HuberAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-88260550480170860882022-07-12T17:05:15.681-05:002022-07-12T17:05:15.681-05:00Thanks for the thoughts! I'll look at the head...Thanks for the thoughts! I'll look at the headers and investigate further.Jo Etzelhttps://www.blogger.com/profile/04277620767760987432noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-11539553599152086372022-07-12T16:17:01.024-05:002022-07-12T16:17:01.024-05:00Hi Jo, the bright ring artifact visible over the b...Hi Jo, the bright ring artifact visible over the brain is from the scalp fat and indicates that the fat suppression pulse was misbehaving. Then, given that the overall images are dark, too, there is the strong suggestion that the RF excitation pulse was also misbehaving. A simple explanation for both phenomena is some sort of erroneous scaling. Usually the transmitter gain for any RF pulse is set based on a calibration step (the current transmitter voltage), usually done invisibly at session start. If that value was corrupted for one entire run - say, it was suddenly reduced by 90% - then all RF pulses would be made weaker. This would lead to ineffective fat suppression and darker images. It would have to be a big factor; 90% or more, probably. <br /><br />Given that you say the effect was constant for an entire run, my suspicion falls on the software or possibly operator error. It's not inconceivable as a hardware error, but it's harder to reconcile with good runs before and after. I'm wondering if either the transmitter reference voltage was saved with a low value for the bum run, or if someone accidentally reset the Tx voltage (in the System tab) for just that run. You may be able to query the transmitter voltage in the DICOM header. If not, pulling one of the DICOMs back onto the scanner and dragging it into the exam queue should reveal what the Tx voltage was.<br /><br />If it happens again and it really hits randomly then you might ask your service engineer to check the RF amplifier in case an attenuator board is getting flaky. But flaky boards tend to produce more intermittent effects. Something affecting one entire scan consistently, and not the scans before or after, that's more likely control/software.<br /><br />praciCal fMRIhttps://www.blogger.com/profile/08592275879035533857noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-27482870761820632462022-03-24T03:16:54.958-05:002022-03-24T03:16:54.958-05:00Thank you! - Vyom (sorry not sure why my name didn...Thank you! - Vyom (sorry not sure why my name didn't show up in the earlier comment)Vyomhttps://www.blogger.com/profile/03203408844942508049noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-58366122756791602052022-03-16T14:47:46.532-05:002022-03-16T14:47:46.532-05:00Sounds interesting, I hope to look at your work mo...Sounds interesting, I hope to look at your work more closely!Jo Etzelhttps://www.blogger.com/profile/04277620767760987432noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-27947309994688448592022-03-16T12:31:24.704-05:002022-03-16T12:31:24.704-05:00We recently published a paper addressing some of t...We recently published a paper addressing some of these same concerns. I started working on this project almost three years ago now (not constantly working on it, moving around etc), and read this blogpost early on in the project. It was really helpful to see your dissatisfaction with the QC-FC metrics and dig deeper into that with the data we had, as well as suggest better positive control measures such as Default Mode Network replicability (compared within same subject across 2 sessions) and identifiability (compared to a template). Thank you for your blog!<br /><br /><br />Neuroinformatics journal: https://link.springer.com/article/10.1007/s12021-022-09565-8<br />biorxiv: https://www.biorxiv.org/content/10.1101/2021.09.18.460908v1.abstractVyomhttps://www.blogger.com/profile/03203408844942508049noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-85322498127701283572021-03-17T15:47:26.111-05:002021-03-17T15:47:26.111-05:00FreeSurfer is definitely a better way to go to pre...FreeSurfer is definitely a better way to go to prepare for analyses. fmriprep and the HCP preprocessing pipelines both use freesurfer for key parts of identifying the surface, so you don't need to use freesurfer directly. It sounds like you don’t need results in the HCP-specific file formats, so I suggest you investigate fmriprep. fmriprep can output functional data as both surface (we use fsaverage5) and volume; the giftis produced by fmriprep can be read by afni/R/etc. for further analysis.Jo Etzelhttps://www.blogger.com/profile/04277620767760987432noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-86630828129987588352021-03-17T15:07:31.872-05:002021-03-17T15:07:31.872-05:00Thank you very much for this great explanation! It...Thank you very much for this great explanation! It's incredibly clear and to the point. <br /><br />I was wondering if you had any experience and/or references regarding the use of FreeSurfer for performing this mapping. I work with functional data aligned with a template space, and a surface atlas (other than HCP); and I would want to perform some further analysis on the mapped cortical surfaces. I saw that there is a function called mri_vol2surf but unfortunately the documentation is not super friendly for a novice like myself. <br /><br />I would appreciate enormously any help possible.Diego Dermannoreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-15938170796425334212021-01-31T23:25:52.141-06:002021-01-31T23:25:52.141-06:00For cross-validation, rather than requiring balanc...For cross-validation, rather than requiring balanced data, you just need stratified sampling, so that the proportions of the various classes in each fold remain approximately constant. In that way, the training folds and each testing fold statistically resemble the overall training data.Anonymoushttps://www.blogger.com/profile/04023224785635536207noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-79720985245238310932021-01-22T13:44:55.899-06:002021-01-22T13:44:55.899-06:00Thanks for the comment! Balance in the training se...Thanks for the comment! Balance in the training set (equal numbers of each class) is *so* important. For fMRI datasets I often suggest subsetting the larger class rather than duplicating members of the smaller class, but I believe both can be appropriate.Jo Etzelhttps://www.blogger.com/profile/04277620767760987432noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-76202207044846030812021-01-22T10:33:32.188-06:002021-01-22T10:33:32.188-06:00Hey, I have come across a crazy scenario. I had 90...Hey, I have come across a crazy scenario. I had 90 datapoints, 45 from class A and 45 from class B. I was doing leave-one-out cross-validation. I was getting 50% training accuracy and 0% test accuracy. I was puzzled like crazy. I started googling and came across your nice post. Great to see that I am not the only one confused about such things. But hey, I figured it out! If the leave-one-out strategy selected the test point from group A, then the training set would consist of 44 points from group A and 45 from group B. Since the labels were completely unpredictable from the data (at least using the the classifier I used), the next best thing it could converge to was to vote for the most prevalent class. So we would get 0% test accuracy, because for every test point the remaining training set would be imbalanced in the opposite way! I was so shocked. Anyway, I have read that there are ways of combating this behaviour, namely, oversamling of the training set to make classes equal before training. Hope this is useful to somebodyAnonymoushttps://www.blogger.com/profile/12889754895726793099noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-6677186663058379642020-10-22T14:25:51.630-05:002020-10-22T14:25:51.630-05:00Sorry for the slow response; I enabled comment mod...Sorry for the slow response; I enabled comment moderation because there has been a lot of spam, but didn't have notification configured properly.<br /><br />For your question (if it still applies!), I'm not sure what you're trying to do, and have never needed the wb_command -cifti-math program. <br /><br />In general, for cifti or wb_command questions the HCP Users google group (https://wiki.humanconnectome.org/pages/viewpage.action?pageId=140509193) may have an answer or be able to help; I suggest you look there.Jo Etzelhttps://www.blogger.com/profile/04277620767760987432noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-78233141312808735142020-09-27T09:10:13.251-05:002020-09-27T09:10:13.251-05:00Dear Pr Jo Etzel,
I need to create the count activ...Dear Pr Jo Etzel,<br />I need to create the count activation map from the task analysis packages downloaded from HCP database, so I didn't undrestand how to work with ‘wb_command -cifti-math’ option, Can you just send me an exemple script to better deal with? TROUDI ABIRhttps://www.blogger.com/profile/13432071962169820656noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-73058051665567558972020-07-27T19:03:00.639-05:002020-07-27T19:03:00.639-05:00Thanks for answering my question. I greatly apprec...Thanks for answering my question. I greatly appreciate it!!!且行https://www.blogger.com/profile/00323966738281993199noreply@blogger.comtag:blogger.com,1999:blog-5737874959005852552.post-89614288994208603172020-07-27T16:57:45.738-05:002020-07-27T16:57:45.738-05:00This is possible - I posted an explanation in http...This is possible - I posted an explanation in http://mvpa.blogspot.com/2020/07/overlaying-in-connectome-workbench-142.htmlJo Etzelhttps://www.blogger.com/profile/04277620767760987432noreply@blogger.com