The more timepoints, the more components, so that aspect seems okay. As far as the new beta MR+FIX training file, I’ll send you a link off list. I would use that.
Matt. From: Yizhou Ma <maxxx...@umn.edu> Date: Thursday, June 20, 2019 at 4:36 PM To: "Glasser, Matthew" <glass...@wustl.edu> Cc: "hcp-users@humanconnectome.org" <hcp-users@humanconnectome.org>, Timothy Hendrickson <hendr...@umn.edu> Subject: Re: [HCP-Users] Multi-run ICA-FIX with excessive movement Dear Matt, I am writing to follow up with this issue. Per your suggestion 2), I looked at the number of signal components in my multi-run ICA-FIX of 20 healthy controls. Again, my data are HCP-style. Each subject has 4 resting state scans (TR=0.8s, length=6.5min each) and 3 task scans (TR=0.8s, length=6min each). We used the HCP preprocessing pipeline. T1 and T2 images were of high quality. Movement was little (typically <1% censored volume by FD_Power < 0.5mm, DVARS Dips looked fine too). We used the standard classifier in FIX. In a multi-run ICA-FIX that concatenated two rfMRI scans (Analysis A): average number of total components: 135.4 ± 34.3; average number of signal components: 11.9 ± 4; average percent of noise components: 90.8 ± 3.6%. In a multi-run that concatenated two fMRI scans and three task scans (Analysis B): average number of total components: 275.2 ± 56.3; average number of signal components: 16.5 ± 4.9; average percent of noise components: 93.8 ± 2.1%. I wasn't able to find any references of previous multirun ICA-FIX results. I looked at 20 subjects in the original HCP, with single-session ICA-FIX for one 15-min rfMRI run: average number of total components: 102 ± 29; average number of signal components: 17.7 ± 4.2; average percent of noise components was 81 ± 6%. Comparing my Analysis A and the original HCP results, it seemed that I had significantly more total components, significantly fewer signal components, and significantly higher percentage of noise components. I have not gone through my components to determine the accuracy of the classfication - that is beyond my expertise at this moment. My question is: do these numbers match with your experience with multi-run ICA-FIX, or do they raise concern about the classfication process? Thank you very much, Cherry On Mon, Apr 22, 2019 at 9:14 PM Yizhou Ma <maxxx...@umn.edu<mailto:maxxx...@umn.edu>> wrote: Great that's quick! On Mon, Apr 22, 2019 at 9:09 PM Glasser, Matthew <glass...@wustl.edu<mailto:glass...@wustl.edu>> wrote: A few weeks maybe if you want a pre-release version. Matt. From: Yizhou Ma <maxxx...@umn.edu<mailto:maxxx...@umn.edu>> Date: Monday, April 22, 2019 at 9:06 PM To: Matt Glasser <glass...@wustl.edu<mailto:glass...@wustl.edu>> Cc: "hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>" <hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>> Subject: Re: [HCP-Users] Multi-run ICA-FIX with excessive movement Thank you Matt. This is really helpful. Any idea when the new classifier you mentioned in 1. will be available? On Mon, Apr 22, 2019 at 8:50 PM Glasser, Matthew <glass...@wustl.edu<mailto:glass...@wustl.edu>> wrote: I guess I haven’t been in the habit of throwing out data like this. Things I would consider would include: 1. MR+FIX classification accuracy (if runs were poorly classified, they won’t be denoised well). I’ll note that we are training an improved MR+FIX classifier using a combination of HCP-YA resting state (single run FIX), HCP-YA task (MR+FIX), and HCP Lifespan (MR+FIX) to address classification issues we have observed with very large numbers of components, subject with very large amounts of motion, and other artifacts that were not a part of the HCP-YA original training data. 2. Unusually small numbers of signal components (though note we found a recent subtle bug whereby if melodic does not finish mixture modeling components, FIX will fail to classify signal components correctly). If there are few signal components this means that either the SNR is very bad or the structured noise has overwhelmed the signal and mixed in too much with the signal, making it hard to separate. 3. DVARS Spikes above baseline (not dips below baseline) in the cleaned timeseries suggest residual noise. I prefer DVARS derived measures to movement tracer derived measures because they tell you something about what is actually happening to the intensities inside the data, whereas movement tracers may be inaccurate reflections of signal intensity fluctuations for a variety of reasons (see Glasser et al 2018 Neuroimage: https://www.sciencedirect.com/science/article/pii/S1053811918303963 for examples). Others in the HCP used different means to identify some of the noise components I mentioned above that weren’t being classified correctly by regular FIX, and might be able to share their suggestions. Matt. From: Yizhou Ma <maxxx...@umn.edu<mailto:maxxx...@umn.edu>> Date: Monday, April 22, 2019 at 8:25 PM To: Matt Glasser <glass...@wustl.edu<mailto:glass...@wustl.edu>> Cc: "hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>" <hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>> Subject: Re: [HCP-Users] Multi-run ICA-FIX with excessive movement Thank you Matt. Do you have some suggestions for the metrics to use to determine scan quality after ICA FIX? Thanks, Cherry On Mon, Apr 22, 2019 at 8:15 PM Glasser, Matthew <glass...@wustl.edu<mailto:glass...@wustl.edu>> wrote: I would decide after cleaning with MR ICA+FIX if you actually have to exclude the scans and run with them all. Matt. From: <hcp-users-boun...@humanconnectome.org<mailto:hcp-users-boun...@humanconnectome.org>> on behalf of Yizhou Ma <maxxx...@umn.edu<mailto:maxxx...@umn.edu>> Date: Monday, April 22, 2019 at 3:47 PM To: "hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>" <hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>> Subject: [HCP-Users] Multi-run ICA-FIX with excessive movement Dear HCP experts, I am writing for a question with multi-run ICA-FIX for my dataset. I have 4 resting state scans (TR=0.8, length=6.5min each) and 3 task scans (TR=0.8, length=6min each) that I intend to run multi-run ICA-FIX on. We used Euclidean norm values to threshold volumes with excessive movement and decided that scans with more than 20% volumes with excessive movement are not usable. I wonder with multi-run ICA-FIX, if it would be problematic to include these scans. In other words, I am trying to decide if I should 1) run multi-run ICA-FIX on scans with less motion, therefore each subject may have different number of scans that are included in multi-run ICA-FIX; or 2) run multi-run ICA-FIX on all scans, and throw out scans with excessive motion afterward. Thank you very much, Cherry _______________________________________________ HCP-Users mailing list HCP-Users@humanconnectome.org<mailto:HCP-Users@humanconnectome.org> http://lists.humanconnectome.org/mailman/listinfo/hcp-users ________________________________ The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail. ________________________________ The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail. ________________________________ The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail. ________________________________ The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail. _______________________________________________ HCP-Users mailing list HCP-Users@humanconnectome.org http://lists.humanconnectome.org/mailman/listinfo/hcp-users
