I hope that everyone had a good time working on this contest! I know that it is over now, but I was just curious about one aspect. I learned that it may be easier to find the presence/absence of the MGMT Promoter Methylation from PET scans (paper).
\nDoes anyone know if it is possible to use some sort of generative network to produce synthetic PET scans from provided MRIs? This seems like a random idea and I know external data would be required, but please let me know if it is possible.
", "messages": [ { "id": "1547657", "postDate": "10/17/2021 13:08:45", "content": "I hope that everyone had a good time working on this contest! I know that it is over now, but I was just curious about one aspect. I learned that it may be easier to find the presence/absence of the MGMT Promoter Methylation from PET scans (paper).
\nDoes anyone know if it is possible to use some sort of generative network to produce synthetic PET scans from provided MRIs? This seems like a random idea and I know external data would be required, but please let me know if it is possible.
", "rawMarkdown": "I hope that everyone had a good time working on this contest! I know that it is over now, but I was just curious about one aspect. I learned that it may be easier to find the presence/absence of the MGMT Promoter Methylation from PET scans ([paper](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6701097/)). \n\nDoes anyone know if it is possible to use some sort of generative network to produce synthetic PET scans from provided MRIs? This seems like a random idea and I know external data would be required, but please let me know if it is possible.", "votes": null }, { "id": "1547841", "postDate": "10/17/2021 16:28:22", "content": "PET in the article is CT, different from MRI.
", "rawMarkdown": "PET in the article is CT, different from MRI.", "votes": null }, { "id": "1547890", "postDate": "10/17/2021 17:23:38", "content": "PET is entirely different than CT or MRI. It is not possible to produce one from the other.
", "rawMarkdown": "PET is entirely different than CT or MRI. It is not possible to produce one from the other.", "votes": null }, { "id": "1548020", "postDate": "10/17/2021 20:44:39", "content": "Indeed, generating synthetic images is possible!
\nThere are approaches that already do generate synthetic PET (for Alzheimer disease) but as far as i know, only for FDG-PET and not FET-PET (which is probably better for accessing methylation than the FDG - mentioned in the article you cited).
\nIf one had some hundred MRI - PET pairs (of the same patient) with this particular disease entity (glioblastoma), it would be possible to train such a system - how it would work out is another question =)
", "rawMarkdown": "Indeed, generating synthetic images is possible!\n\nThere are approaches that already do generate synthetic PET (for Alzheimer disease) but as far as i know, only for FDG-PET and not FET-PET (which is probably better for accessing methylation than the FDG - mentioned in the article you cited).\n\nIf one had some hundred MRI - PET pairs (of the same patient) with this particular disease entity (glioblastoma), it would be possible to train such a system - how it would work out is another question =)", "votes": null }, { "id": "1548025", "postDate": "10/17/2021 21:07:42", "content": "Hi dr. Konya,
\nThanks for the reply. This seems really interesting. I dont have too much experience working with PET scans, so thanks for letting me know what would work well for glioblastomas. Do you happen to know if ther are any open source datasets with MRI/FDG-PET pairs for the alzheimers disease you were mentioning?
", "rawMarkdown": "Hi dr. Konya,\n\nThanks for the reply. This seems really interesting. I dont have too much experience working with PET scans, so thanks for letting me know what would work well for glioblastomas. Do you happen to know if ther are any open source datasets with MRI/FDG-PET pairs for the alzheimers disease you were mentioning?", "votes": null }, { "id": "1548029", "postDate": "10/17/2021 21:15:09", "content": "@saumandas i think this article is a good starting point:
\nhttps://braininformatics.springeropen.com/articles/10.1186/s40708-020-00104-2
\nin there this dataset is mentioned: http://adni.loni.usc.edu/
Thank you so much!
", "rawMarkdown": "Thank you so much!", "votes": null }, { "id": "1548047", "postDate": "10/17/2021 21:51:29", "content": "I am very skeptical of the approach of using GANs to generate synthetic PET scans from MR images. I suspect it's more theorycraft than clinical. It isn't possible to extract the data that a PET scan produces from an MR. Any data created would be made to mimic a PET, but it would not actually contain data from the MR itself.
", "rawMarkdown": "I am very skeptical of the approach of using GANs to generate synthetic PET scans from MR images. I suspect it's more theorycraft than clinical. It isn't possible to extract the data that a PET scan produces from an MR. Any data created would be made to mimic a PET, but it would not actually contain data from the MR itself.", "votes": null } ], "comments": [ { "id": 1547841, "author_name": "drtausamaru", "author_url": "", "post_date": "10/17/2021 16:28:22", "content": "PET in the article is CT, different from MRI.
", "votes": null, "replies": [] }, { "id": 1547890, "author_name": "davidbroberts", "author_url": "", "post_date": "10/17/2021 17:23:38", "content": "PET is entirely different than CT or MRI. It is not possible to produce one from the other.
", "votes": null, "replies": [] }, { "id": 1548020, "author_name": "sandorkonya", "author_url": "", "post_date": "10/17/2021 20:44:39", "content": "Indeed, generating synthetic images is possible!
\nThere are approaches that already do generate synthetic PET (for Alzheimer disease) but as far as i know, only for FDG-PET and not FET-PET (which is probably better for accessing methylation than the FDG - mentioned in the article you cited).
\nIf one had some hundred MRI - PET pairs (of the same patient) with this particular disease entity (glioblastoma), it would be possible to train such a system - how it would work out is another question =)
", "votes": null, "replies": [ { "id": 1548025, "author_name": "saumandas", "author_url": "", "post_date": "10/17/2021 21:07:42", "content": "Hi dr. Konya,
\nThanks for the reply. This seems really interesting. I dont have too much experience working with PET scans, so thanks for letting me know what would work well for glioblastomas. Do you happen to know if ther are any open source datasets with MRI/FDG-PET pairs for the alzheimers disease you were mentioning?
", "votes": null, "replies": [] }, { "id": 1548029, "author_name": "sandorkonya", "author_url": "", "post_date": "10/17/2021 21:15:09", "content": "@saumandas i think this article is a good starting point:
\nhttps://braininformatics.springeropen.com/articles/10.1186/s40708-020-00104-2
\nin there this dataset is mentioned: http://adni.loni.usc.edu/
Thank you so much!
", "votes": null, "replies": [] }, { "id": 1548047, "author_name": "davidbroberts", "author_url": "", "post_date": "10/17/2021 21:51:29", "content": "I am very skeptical of the approach of using GANs to generate synthetic PET scans from MR images. I suspect it's more theorycraft than clinical. It isn't possible to extract the data that a PET scan produces from an MR. Any data created would be made to mimic a PET, but it would not actually contain data from the MR itself.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1547657": "I hope that everyone had a good time working on this contest! I know that it is over now, but I was just curious about one aspect. I learned that it may be easier to find the presence/absence of the MGMT Promoter Methylation from PET scans ([paper](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6701097/)). \n\nDoes anyone know if it is possible to use some sort of generative network to produce synthetic PET scans from provided MRIs? This seems like a random idea and I know external data would be required, but please let me know if it is possible.", "1547841": "PET in the article is CT, different from MRI.", "1547890": "PET is entirely different than CT or MRI. It is not possible to produce one from the other.", "1548020": "Indeed, generating synthetic images is possible!\n\nThere are approaches that already do generate synthetic PET (for Alzheimer disease) but as far as i know, only for FDG-PET and not FET-PET (which is probably better for accessing methylation than the FDG - mentioned in the article you cited).\n\nIf one had some hundred MRI - PET pairs (of the same patient) with this particular disease entity (glioblastoma), it would be possible to train such a system - how it would work out is another question =)", "1548025": "Hi dr. Konya,\n\nThanks for the reply. This seems really interesting. I dont have too much experience working with PET scans, so thanks for letting me know what would work well for glioblastomas. Do you happen to know if ther are any open source datasets with MRI/FDG-PET pairs for the alzheimers disease you were mentioning?", "1548029": "saumandas i think this article is a good starting point:\nhttps://braininformatics.springeropen.com/articles/10.1186/s40708-020-00104-2\nin there this dataset is mentioned: http://adni.loni.usc.edu/", "1548035": "Thank you so much!", "1548047": "I am very skeptical of the approach of using GANs to generate synthetic PET scans from MR images. I suspect it's more theorycraft than clinical. It isn't possible to extract the data that a PET scan produces from an MR. Any data created would be made to mimic a PET, but it would not actually contain data from the MR itself." }, "source": "meta" }