 |
 |
John, Thank you so much for taking the time to respond. I can tell you that the template does not have any 0 voxels in it, but I'm not sure what methods were used to achieve that (it was created by another research group). The better strategy would have been to manually edit all the scans and then make the template from those (as you did when creating the stock SPM TPM), but that also takes more time. The TPMs seem to work very well for segmentation in SPM but they have not been extensively evaluated for Dartel registration purposes. The TPMs also come with an atlas that was "meticulously hand edited" after being warped to template space. I had hoped that using these to calculate tissue volumes from a sample of patients with varying degrees of Alzheimer's disease would be more accurate than simply using existing TPMs/atlases that do not match the population as well and have not been cleaned up by manual intervention. These issues could be avoided by using the manually edited TPMs for unified segmentation and then using another strategy (SyN or a pairwise algorithm) to propagate the atlas to individual space, but I am an SPM fan and had hoped to stay within an SPM/Dartel framework for the project. Are both of the ideas I proposed below suboptimal or might one of those methods be worth a try? I can't think of a better way to get the atlas into individual space within an SPM framework (obviously the deformations from Unified Segmentation could be used, but I worry about lower registration accuracy because it does not have as many parameters to work with as Dartel). Peter ________________________________ From: John Ashburner <[log in to unmask]> Sent: Friday, January 11, 2019 4:06 PM To: Stewart, Peter Cc: SPM Subject: [External] Re: [SPM] DARTEL experts: can you create a set of templates from a final template by smoothing? Be careful with hand editing templates built by Dartel. Dartel alignment is driven by the gradients of the logarithms of the values in the templates. Hand editing has probably changed some of the values to zero, which could potentially mess things up. The use of logs comes from the assumption that the template encodes the mean of a categorical distribution (https://en.wikipedia.org/wiki/Categorical_distribution<https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fen.wikipedia.org%2Fwiki%2FCategorical_distribution&data=02%7C01%7Cpstewart1%40geisinger.edu%7C011956067c4b4cc5040808d67808a8f7%7C37d46c567c664402a16055c2313b910d%7C0%7C0%7C636828375923863132&sdata=Hpr%2BxC1pXZ1ZUbsO9GhqmYyNvI%2FUvoHtiajsj%2FAlUY8%3D&reserved=0>). A template GM voxel of exactly zero is basically saying that there should never ever ever be any trace of GM in that location, so the algorithm will totally freak out and break if a GM voxel does occur there. Best regards, -John On Fri, 11 Jan 2019 at 19:01, Stewart, Peter <[log in to unmask]<mailto:[log in to unmask]>> wrote: Hi Everybody, I have a question and I would very much appreciate input from DARTEL experts. The short version is: I have a hand edited TPM (it was created by registering a set of 202 individuals using DARTEL and then hand editing the final template to enhance accuracy and eliminate extraneous voxels). I would like to warp some new scans into the TPM space. Given that the final template is hand edited, I’m unsure how to proceed (i.e., using the set of templates 1-6 would not correspond exactly to the final TPM, because they are not hand edited). One idea I had was using the templates created en route to the hand edited TPM as the first templates, and then selecting the TPMs as the final template for the registration. This would hopefully have the increased robustness benefits of using a full set of templates while getting everything into the right space with the final step. The other idea was to create templates 1-5 (or actually 2-4, if I recall you only need 5 for Run Dartel: Add to Existing Template) via smoothing the TPMs by kernels of varying sizes to make progressively more blurred images (essentially the reverse of what DARTEL seeks to do, by making increasingly crisp templates). I’m happy to provide more context if it would be helpful. Any ideas on which approach might be better, if there is any difference, and if it’s the second approach, what factors might I want to smooth by? I greatly appreciate your comments, so thank you in advance. Peter Peter V. Stewart, Psy.D. | Staff Neuropsychologist | Assistant Program Director, Neuropsychology Postdoctoral Training | Neuropsychology Internship Coordinator | Geisinger Health System | [log in to unmask]<mailto:[log in to unmask]> [cid:image002.png@01D190C9.4E647E10] ________________________________ IMPORTANT WARNING: The information in this message (and the documents attached to it, if any) is confidential and may be legally privileged. It is intended solely for the addressee. Access to this message by anyone else is unauthorized. If you are not the intended recipient, any disclosure, copying, distribution or any action taken, or omitted to be taken, in reliance on it is prohibited and may be unlawful. If you have received this message in error, please delete all electronic copies of this message (and the documents attached to it, if any), destroy any hard copies you may have created and notify me immediately by replying to this email. Thank you. Geisinger Health System utilizes an encryption process to safeguard Protected Health Information and other confidential data contained in external e-mail messages. If email is encrypted, the recipient will receive an e-mail instructing them to sign on to the Geisinger Health System Secure E-mail Message Center to retrieve the encrypted e-mail. -- Prof John Ashburner Professor of Imaging Science UCL Institute of Neurology Queen Square Wellcome Trust Centre for Neuroimaging University College London 12 Queen Square, London, WC1N 3BG E: [log in to unmask]<mailto:[log in to unmask]> T: +44 (0)20 3448 4365 http://www.fil.ion.ucl.ac.uk/<https://na01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.fil.ion.ucl.ac.uk%2F&data=02%7C01%7Cpstewart1%40geisinger.edu%7C011956067c4b4cc5040808d67808a8f7%7C37d46c567c664402a16055c2313b910d%7C0%7C0%7C636828375923863132&sdata=yg9RpfYdyYk%2BPzslgAw8Xs1VM1Gc9c2m3LHmcYGZRR0%3D&reserved=0>