open-discussion
open-discussion > RE: Too many gradients excluded by "default"
Apr 20, 2017 02:04 AM | minchulusa
RE: Too many gradients excluded by "default"
Hi Martin
Thank you a lot for your kind advice, it was helpful.
I cameout with additional questions on your answer.
1) Upon your answer "Re 2)" , I want to set the tolerance threshold to whether create or not created final QCed DWI and DTIs.
I looked up at media wiki, but I cannot be sure which Parameter is for tolerance setting.
I have two guesses and marked "1" and "2" in attached image.
Please let me know the function and meanings of those two parameters.
2) Another question is that for "Re 3)", Am I supposed to train DTIPrep with my own artifact free data? Because I though DTIPrep has
kind of built-in artifact free samples and compare with my data.
Thanks in advance and look forward to listen.
Minchul
Originally posted by Martin Styner:
Thank you a lot for your kind advice, it was helpful.
I cameout with additional questions on your answer.
1) Upon your answer "Re 2)" , I want to set the tolerance threshold to whether create or not created final QCed DWI and DTIs.
I looked up at media wiki, but I cannot be sure which Parameter is for tolerance setting.
I have two guesses and marked "1" and "2" in attached image.
Please let me know the function and meanings of those two parameters.
2) Another question is that for "Re 3)", Am I supposed to train DTIPrep with my own artifact free data? Because I though DTIPrep has
kind of built-in artifact free samples and compare with my data.
Thanks in advance and look forward to listen.
Minchul
Originally posted by Martin Styner:
Hi
Minchul
Re 1) the first denoising/smoothing option DENOISING_bCheck smooth the DWI data prior to any check which will smooth out any minor artifacts, which is why the previously rejected DWIs are now considered fine. It's "okay" to use that option, though we usually do not (I prefer un-smoothed data, but other researchers prefer smoothed data). This would definitely be something that you would need to report in any papers resulting from your work. The options JOINDENOISING_bCheck, BRAINMAST_bCheck, and DOMINANTDIRECTION_bCheck have no effect on the DWI cleaning/rejections.
What we usually do when too many DWI get rejected despite appearing appropriate is to change the thresholds for SLICE and INTERLACE checks. The report txt and xml files give you information which step rejects the most DWI and why.
Re 2) That depends on your settings. You can tell DTIPrep not to create the final QCed DWI and DTI images when the resulting DWI has more than a certain percentage of rejected DWIs (tolerance threshold). At the most permissive (when data is very noisy), we usually do not go beyond 40% rejection, while our usual studies use 20-30% rejection tolerance. 50% seems to high to me and I would NOT use that data for further analysis.
re 3)DOMINANTDIRECTION_bCheck = vibration artifact check. This is not commonly needed unless you have earlier Tim Trio data or suspect presence of vibration artifacts. The issue here is that you need to train the classifier by running the check on sample data that looks good (the xml contains the entropy results, see our papers on this topic) and then use the mean & stdev from that sample data in the protocol to identify the presence or absence of such an artifact.
re 3) simulation based bias analysis: unfortunately that is not in DTIPrep yet. We simply don't have the manpower to put it in (there is no grant that support DTIPrep and we are currently only maintaining it and do not add any major functionality). This would be great as it would allow to reject/accept DTI data based on the expected error rather than a simple tolerance threshold.
re 4) DTIPrep offers support for multiple brain masking options. If you want to use that option you need to point DTIPrep to it (it's part of Slicer). We commonly suggest the use of FSL's bet/bet2 with IDWI (double click on the protocol option and DTIPrep will allow you to select which masking option to choose)
Martin
Originally posted by Minchul Kim:
Re 1) the first denoising/smoothing option DENOISING_bCheck smooth the DWI data prior to any check which will smooth out any minor artifacts, which is why the previously rejected DWIs are now considered fine. It's "okay" to use that option, though we usually do not (I prefer un-smoothed data, but other researchers prefer smoothed data). This would definitely be something that you would need to report in any papers resulting from your work. The options JOINDENOISING_bCheck, BRAINMAST_bCheck, and DOMINANTDIRECTION_bCheck have no effect on the DWI cleaning/rejections.
What we usually do when too many DWI get rejected despite appearing appropriate is to change the thresholds for SLICE and INTERLACE checks. The report txt and xml files give you information which step rejects the most DWI and why.
Re 2) That depends on your settings. You can tell DTIPrep not to create the final QCed DWI and DTI images when the resulting DWI has more than a certain percentage of rejected DWIs (tolerance threshold). At the most permissive (when data is very noisy), we usually do not go beyond 40% rejection, while our usual studies use 20-30% rejection tolerance. 50% seems to high to me and I would NOT use that data for further analysis.
re 3)DOMINANTDIRECTION_bCheck = vibration artifact check. This is not commonly needed unless you have earlier Tim Trio data or suspect presence of vibration artifacts. The issue here is that you need to train the classifier by running the check on sample data that looks good (the xml contains the entropy results, see our papers on this topic) and then use the mean & stdev from that sample data in the protocol to identify the presence or absence of such an artifact.
re 3) simulation based bias analysis: unfortunately that is not in DTIPrep yet. We simply don't have the manpower to put it in (there is no grant that support DTIPrep and we are currently only maintaining it and do not add any major functionality). This would be great as it would allow to reject/accept DTI data based on the expected error rather than a simple tolerance threshold.
re 4) DTIPrep offers support for multiple brain masking options. If you want to use that option you need to point DTIPrep to it (it's part of Slicer). We commonly suggest the use of FSL's bet/bet2 with IDWI (double click on the protocol option and DTIPrep will allow you to select which masking option to choose)
Martin
Originally posted by Minchul Kim:
Hello ,
My name is MC kim, and I need some help.
The DTIPrep version I got is 1.2.4.
I try to preprocess my DWI images with DTIPrep, and I got some questions in doing it.
1) I initially tried with 'default' protocol, and DTIPrep excluded 17 out for 31 gradients from a DWI data set from a subject.
the reason was due to interlace artifact. When I visual checked the DWI data, I couldn't agree that much artifact.
I attached the captured QC result file.
I tried preprocessing again, and this time I changed the protocol, instead of using 'default'.
By default, DENOISING_bCheck, JOINDENOISING_bCheck, BRAINMAST_bCheck, and DOMINANTDIRECTION_bCheck at setted "No", and I changed settings into "YES"
As I tried again with all "YES" all of the directions were included in QCed DWI.
Thus I though by changing protocols to all "YES", the protocol became more tolerant to artifact.
My question is is it okay to run with the protocol setting of all "YES"? Is there any potential problem?
2) Even though the "default"protocol excluded more than half of DWI gradient directions, DTIPrep still generated DTI images such as FA, MD, etc.
Can I use this DTI image to analysis or tract generation? or should I prepration again with different protocol setting.
3)I read that DTIPrep has two nice functions, those are "vibration artifact checking" and "simulation based bias analysis".
I want to know which parameter on protocol setting represents those functions.
My guess is "DOMINANTDIRECTION_bCheck" maybe vibration artifact correction, am I right?
4) When I click on "Default" protocol, I get the error message saying "The following programs have not been found. Please enter the path manually if necessary> DiffusionWeightedVolumeMasking"
Can you tell me where can I get this program or how can I solve the problem?
Thanks in advance and I look forward to hear from you.
MC Kim
My name is MC kim, and I need some help.
The DTIPrep version I got is 1.2.4.
I try to preprocess my DWI images with DTIPrep, and I got some questions in doing it.
1) I initially tried with 'default' protocol, and DTIPrep excluded 17 out for 31 gradients from a DWI data set from a subject.
the reason was due to interlace artifact. When I visual checked the DWI data, I couldn't agree that much artifact.
I attached the captured QC result file.
I tried preprocessing again, and this time I changed the protocol, instead of using 'default'.
By default, DENOISING_bCheck, JOINDENOISING_bCheck, BRAINMAST_bCheck, and DOMINANTDIRECTION_bCheck at setted "No", and I changed settings into "YES"
As I tried again with all "YES" all of the directions were included in QCed DWI.
Thus I though by changing protocols to all "YES", the protocol became more tolerant to artifact.
My question is is it okay to run with the protocol setting of all "YES"? Is there any potential problem?
2) Even though the "default"protocol excluded more than half of DWI gradient directions, DTIPrep still generated DTI images such as FA, MD, etc.
Can I use this DTI image to analysis or tract generation? or should I prepration again with different protocol setting.
3)I read that DTIPrep has two nice functions, those are "vibration artifact checking" and "simulation based bias analysis".
I want to know which parameter on protocol setting represents those functions.
My guess is "DOMINANTDIRECTION_bCheck" maybe vibration artifact correction, am I right?
4) When I click on "Default" protocol, I get the error message saying "The following programs have not been found. Please enter the path manually if necessary> DiffusionWeightedVolumeMasking"
Can you tell me where can I get this program or how can I solve the problem?
Thanks in advance and I look forward to hear from you.
MC Kim
Threaded View
| Title | Author | Date |
|---|---|---|
| minchulusa | Apr 19, 2017 | |
| Martin Styner | Apr 19, 2017 | |
| minchulusa | Apr 20, 2017 | |
| Martin Styner | Apr 20, 2017 | |
| minchulusa | Apr 20, 2017 | |
| Martin Styner | Apr 20, 2017 | |
| minchulusa | May 11, 2017 | |
| minchulusa | May 18, 2017 | |
| Martin Styner | May 18, 2017 | |
| Martin Styner | May 11, 2017 | |