![\"img\"](\"https://i.imgur.com/ct9Ea5J.gif\")
My first impression about this competition was that we need to classify a given ID to have/not have a brain tumor. My intuition was that the presence of MGMT promoter methylation leads to a brain tumor and that MRI sequences without MGMT promoters do not have a brain tumor.
\nI had logged the entire dataset onto W&B Tables. You can find more information about this in this discussion post.
\n
After spending numerous hours looking through the dataset, I got confused because I could see the tumor-like artifact in almost all sequences, irrespective of the MGMT value.
\nGlioblastoma is the most frequent malignant primary tumor in the brain. It has a very poor prognosis, with a median survival of less than a year. The current standard of care consists of surgical resection followed by radiotherapy in addition to alkylating chemotherapy with temozolomide.
\n\n\n📍 Alkylating agents are compounds that work by adding an alkyl group to the guanine base of the DNA molecule, preventing the double helix strands from linking as they should. This causes breakage of the DNA strands, affecting the ability of the cancer cell to multiply.
\n
MGMT (O[6]-methylguanine-DNA methyltransferase) is a DNA repair enzyme. This enzyme rescues tumor cells from alkylating agent-induced damage, leading to chemotherapy resistance with alkylating agents.
\nEpigenetic silencing of the MGMT gene by promoter methylation results in decreased MGMT protein expression which potentially increases therapy sensitivity.
\n\n\n📍 Epigenetic silencing is a natural phenomenon in which the expression of genes is regulated through modifications of DNA, RNA.
\n
This is based on this article.
\n\"The O6-methylguanine-DNA methyl-transferase (MGMT) gene is located on chromosome 10q26.3\" - obviously we can't see MGMT in the MRI sequences.
\"Importantly, epigenetic modification could silence the MGMT gene, commonly through methylation of the cytosine-phosphate-guanine (CpG) island at specific CpG sites within the MGMT gene promoter.\": Epigenetic modification (silencing) happens naturally. This methylation of MGMT gene promoter induces loss/low levels of functional MGMT protein due to which alkylating chemotherapy works on the patient.
\"Methylation of the MGMT gene promoter has been observed in approximately 50% of grade IV gliomas, commonly referred to as glioblastoma multiforme\": This somewhat feels consistent with the distribution of the labels in the training set.
\"Originally, the first evidence that associated MGMT methylation status and response to TMZ emerged from the ‘Stupp trial’ \": Here's the paper on Stupp trial. From the conclusion section of this paper: \" Patients with glioblastoma containing a methylated MGMT promoter benefited from temozolomide, whereas those who did not have a methylated MGMT promoter did not have such a benefit. \"
![\"img\"](\"https://i.imgur.com/7M3O6Lh.jpg\")
\n(Source)
From the conclusion section of, \"Open Access\nClinically Relevant Imaging Features for MGMT Promoter Methylation in Multiple Glioblastoma Studies: A Systematic Review and Meta-Analysis\"
\n\n\nMGMT promotor methylated glioblastoma is likely to show less aggressive imaging features than MGMT promotor unmethylated glioblastoma.
\n
I think the paper titled, \"Structural and advanced imaging in predicting MGMT promoter methylation of primary glioblastoma: a region of interest based analysis finds the correlation between image feature and presence of MGMT.
\nI didn't get into the details of the paper, but from a quick scan I think that there are some minute differences in the tumor with and without MGMT.
\n![\"img\"](\"https://i.imgur.com/paoR9fM.png\")
\n\nMy thoughts: Given deep learning specializes in extracting \"meaningful\" features, let's leave it to the models to do the job and focus on using good ML practices. Domain knowledge will be super helpful in pre-/post-processing.
\n
I hope it might help clear some confusion.
", "messages": [ { "id": 1470454, "postDate": "2021-08-13T14:20:25.910Z", "content": "My first impression about this competition was that we need to classify a given ID to have/not have a brain tumor. My intuition was that the presence of MGMT promoter methylation leads to a brain tumor and that MRI sequences without MGMT promoters do not have a brain tumor.
\nI had logged the entire dataset onto W&B Tables. You can find more information about this in this discussion post.
\n
After spending numerous hours looking through the dataset, I got confused because I could see the tumor-like artifact in almost all sequences, irrespective of the MGMT value.
\nGlioblastoma is the most frequent malignant primary tumor in the brain. It has a very poor prognosis, with a median survival of less than a year. The current standard of care consists of surgical resection followed by radiotherapy in addition to alkylating chemotherapy with temozolomide.
\n\n\n📍 Alkylating agents are compounds that work by adding an alkyl group to the guanine base of the DNA molecule, preventing the double helix strands from linking as they should. This causes breakage of the DNA strands, affecting the ability of the cancer cell to multiply.
\n
MGMT (O[6]-methylguanine-DNA methyltransferase) is a DNA repair enzyme. This enzyme rescues tumor cells from alkylating agent-induced damage, leading to chemotherapy resistance with alkylating agents.
\nEpigenetic silencing of the MGMT gene by promoter methylation results in decreased MGMT protein expression which potentially increases therapy sensitivity.
\n\n\n📍 Epigenetic silencing is a natural phenomenon in which the expression of genes is regulated through modifications of DNA, RNA.
\n
This is based on this article.
\n\"The O6-methylguanine-DNA methyl-transferase (MGMT) gene is located on chromosome 10q26.3\" - obviously we can't see MGMT in the MRI sequences.
\"Importantly, epigenetic modification could silence the MGMT gene, commonly through methylation of the cytosine-phosphate-guanine (CpG) island at specific CpG sites within the MGMT gene promoter.\": Epigenetic modification (silencing) happens naturally. This methylation of MGMT gene promoter induces loss/low levels of functional MGMT protein due to which alkylating chemotherapy works on the patient.
\"Methylation of the MGMT gene promoter has been observed in approximately 50% of grade IV gliomas, commonly referred to as glioblastoma multiforme\": This somewhat feels consistent with the distribution of the labels in the training set.
\"Originally, the first evidence that associated MGMT methylation status and response to TMZ emerged from the ‘Stupp trial’ \": Here's the paper on Stupp trial. From the conclusion section of this paper: \" Patients with glioblastoma containing a methylated MGMT promoter benefited from temozolomide, whereas those who did not have a methylated MGMT promoter did not have such a benefit. \"
\n(Source)
From the conclusion section of, \"Open Access\nClinically Relevant Imaging Features for MGMT Promoter Methylation in Multiple Glioblastoma Studies: A Systematic Review and Meta-Analysis\"
\n\n\nMGMT promotor methylated glioblastoma is likely to show less aggressive imaging features than MGMT promotor unmethylated glioblastoma.
\n
I think the paper titled, \"Structural and advanced imaging in predicting MGMT promoter methylation of primary glioblastoma: a region of interest based analysis finds the correlation between image feature and presence of MGMT.
\nI didn't get into the details of the paper, but from a quick scan I think that there are some minute differences in the tumor with and without MGMT.
\n
\n\nMy thoughts: Given deep learning specializes in extracting \"meaningful\" features, let's leave it to the models to do the job and focus on using good ML practices. Domain knowledge will be super helpful in pre-/post-processing.
\n
I hope it might help clear some confusion.
", "rawMarkdown": "My first impression about this competition was that we need to classify a given ID to have/not have a brain tumor. My intuition was that the presence of MGMT promoter methylation leads to a brain tumor and that MRI sequences without MGMT promoters do not have a brain tumor.\n\n## I was wrong. How?\n\nI had logged the entire dataset onto W&B Tables. You can find more information about this in this [discussion post](http://wandb.me/kaggle-256051). \n\n#### [Check out the logged Tables here ▶️](http://wandb.me/brain-tumor-tables)\n\n\n\nAfter spending numerous hours looking through the dataset, I got confused because I could see the tumor-like artifact in almost all sequences, irrespective of the MGMT value. \n\n## So what's MGMT?\n\nGlioblastoma is the most frequent malignant primary tumor in the brain. It has a very poor prognosis, with a median survival of less than a year. The current standard of care consists of surgical resection followed by radiotherapy in addition to **alkylating chemotherapy** with temozolomide.\n\n> 📍 Alkylating agents are compounds that work by adding an alkyl group to the guanine base of the DNA molecule, preventing the double helix strands from linking as they should. This causes breakage of the DNA strands, affecting the ability of the cancer cell to multiply.\n\nMGMT (O[6]-methylguanine-DNA methyltransferase) is a DNA repair enzyme. This enzyme rescues tumor cells from alkylating agent-induced damage, leading to chemotherapy resistance with alkylating agents. \n\n**Epigenetic silencing of the MGMT gene by promoter methylation** results in decreased MGMT protein expression which potentially increases therapy sensitivity.\n\n> 📍 Epigenetic silencing is a natural phenomenon in which the expression of genes is regulated through modifications of DNA, RNA.\n\n## Update: More on MGMT's connection with Glioblastoma \n\nThis is based on this [article](https://oncologypro.esmo.org/education-library/factsheets-on-biomarkers/mgmt-promoter-methylation-in-glioma).\n\n* \"The O6-methylguanine-DNA methyl-transferase (MGMT) gene is located on chromosome 10q26.3\" - obviously we can't see MGMT in the MRI sequences. \n\n* \"Importantly, epigenetic modification could silence the MGMT gene, commonly through methylation of the cytosine-phosphate-guanine (CpG) island at specific CpG sites within the MGMT gene promoter.\": Epigenetic modification (silencing) happens naturally. This methylation of MGMT gene promoter induces loss/low levels of functional MGMT protein due to which alkylating chemotherapy works on the patient.\n\n* \"Methylation of the MGMT gene promoter has been observed in approximately 50% of grade IV gliomas, commonly referred to as glioblastoma multiforme\": This somewhat feels consistent with the distribution of the labels in the training set. \n\n* \"Originally, the first evidence that associated MGMT methylation status and response to TMZ emerged from the ‘Stupp trial’ \": Here's the [paper on Stupp trial](https://pubmed.ncbi.nlm.nih.gov/15758010/). From the conclusion section of this paper: \" Patients with glioblastoma containing a methylated MGMT promoter benefited from temozolomide, whereas those who did not have a methylated MGMT promoter did not have such a benefit. \"\n\n\n([Source](https://oncologypro.esmo.org/var/esmo/storage/images/media/images-op/biomarkers/mgmt-promoter-methylation-in-glioma/long-term-survival-results-according-to-type-of-treatment-and-mgmt-status-in-the-stupp-trial/3328288-1-eng-GB/Long-term-survival-results-according-to-type-of-treatment-and-MGMT-status-in-the-Stupp-trial_i1200.jpg))\n\n## MGMT Testing Methods\n\n* \"To date, no large consensus exists on the optimal method of assessment of MGMT gene promoter methylation, and different assays are available for this purpose. Regardless of the type of the chosen technique, the preliminary phases consist of genomic DNA extraction, it's qualitative/quantitative check, and bisulfite conversion.\": This process must take time and involve surgery to extract DNA. If an MRI sequence can help detect MGMT that would be great. \n\n## MRI and MGMT Connection\n\nFrom the conclusion section of, [\"Open Access\nClinically Relevant Imaging Features for MGMT Promoter Methylation in Multiple Glioblastoma Studies: A Systematic Review and Meta-Analysis\"](http://www.ajnr.org/content/early/2018/07/12/ajnr.A5711)\n\n> MGMT promotor methylated glioblastoma is likely to show less aggressive imaging features than MGMT promotor unmethylated glioblastoma.\n\nI think the paper titled, [\"Structural and advanced imaging in predicting MGMT promoter methylation of primary glioblastoma: a region of interest based analysis](https://bmccancer.biomedcentral.com/articles/10.1186/s12885-018-4114-2) finds the correlation between image feature and presence of MGMT. \n\nI didn't get into the details of the paper, but from a quick scan I think that there are some minute differences in the tumor with and without MGMT.\n\n\n\n> My thoughts: Given deep learning specializes in extracting \"meaningful\" features, let's leave it to the models to do the job and focus on using good ML practices. Domain knowledge will be super helpful in pre-/post-processing.\n\nI hope it might help clear some confusion. ", "votes": 56 }, { "id": 1481279, "postDate": "2021-08-19T12:09:15.567Z", "content": "My super (overly)simple understanding is that MGMT promoter methylation is strongly correlated with chemotherapy success.
\nTherefore.
\nWe are classifying whether a tumour is likely to respond to chemotherapy or not.
", "rawMarkdown": "My super (overly)simple understanding is that MGMT promoter methylation is strongly correlated with chemotherapy success.\n\nTherefore.\n\n**We are classifying whether a tumour is likely to respond to chemotherapy or not.**", "votes": 6, "replies": [ { "id": 1489087, "postDate": "2021-08-24T17:40:36.023Z", "content": "Agreed. I guess by now most of us are certain about the problem statement.
", "rawMarkdown": "Agreed. I guess by now most of us are certain about the problem statement. \n\n" } ] }, { "id": 1472631, "postDate": "2021-08-15T02:50:31.837Z", "content": "I've looked at hundreds of series in both sets and have only seen a couple studies that I could not immediately identify a tumor in.
\nI suspect the ones I didn't see tumors on were because they were either small, not imaged well, better identified on a different series .. or I just didn't recognize it. Damnit Jim, I'm a tech not a doctor!.
\nMy assumptions are:
\n\n\nOur task is to determine if there is some difference between glioblastomas with MGMT and those without.
\n
I totally agree with this and my quick research reflects that. I have updated the discussion post with more info and useful links. It might be useful.
", "rawMarkdown": "> Our task is to determine if there is some difference between glioblastomas with MGMT and those without.\n\nI totally agree with this and my quick research reflects that. I have updated the discussion post with more info and useful links. It might be useful." }, { "id": 1503603, "postDate": "2021-09-05T14:14:59.710Z", "content": "As far as I understand, methylation is determined only by molecular analysis, for which you need to cut out a piece of the tumor. in those articles that I read, it is argued that at the moment no differences have been found between methylated and unmethylated tumors on MRI, because the process takes place at the DNA level. methylation - blocking the work of the mgmt gene,
\nwhich restores cancer cells during chemotherapy. Another option to find out that the gene is methylated is to look at MRI scans taken at different times - before and after the start of chemotherapy, which is impossible given that we do not have such data.
Thanks for this, Initially I also thought of some correlation between tumour and MGMT, but after looking at many images I concluded I might be wrong.
", "rawMarkdown": "Thanks for this, Initially I also thought of some correlation between tumour and MGMT, but after looking at many images I concluded I might be wrong.", "votes": 1, "replies": [ { "id": 1470615, "postDate": "2021-08-13T16:11:30.840Z", "content": "Just to confirm, there are tumors in all scans, and we are predicting whether there is MGMT promoter methylation, right?
", "rawMarkdown": "Just to confirm, there are tumors in all scans, and we are predicting whether there is MGMT promoter methylation, right?" }, { "id": 1470626, "postDate": "2021-08-13T16:17:18.370Z", "content": "Tumours in all scans? I don't think so, they are random I guess.
\nI am not sure though
I think so. I am still yet to understand the problem completely. See this: https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/252843#1441395
", "rawMarkdown": "I think so. I am still yet to understand the problem completely. See this: https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/252843#1441395", "votes": 1 }, { "id": 1470640, "postDate": "2021-08-13T16:26:48.793Z", "content": "wow,thanks for this!
", "rawMarkdown": "wow,thanks for this!\n", "votes": 1 }, { "id": 1470702, "postDate": "2021-08-13T17:02:41.627Z", "content": "I think we have a brain tumor in every scan. If you visit the table and just look through the scans, it's easy to find lesions/edema/tumorish stuff.
", "rawMarkdown": "I think we have a brain tumor in every scan. If you visit the table and just look through the scans, it's easy to find lesions/edema/tumorish stuff.", "votes": 5 }, { "id": 1471559, "postDate": "2021-08-14T09:40:32.667Z", "content": "My understanding is for those classify 0, you may or may not have tumour. For those with class 1, you definitely have tumour. The level of presence of MGMT is the factor deciding class 0 or 1. The question I am thinking now is what is the features contribute to class 1 or 0. Anyone have any idea?
", "rawMarkdown": "My understanding is for those classify 0, you may or may not have tumour. For those with class 1, you definitely have tumour. The level of presence of MGMT is the factor deciding class 0 or 1. The question I am thinking now is what is the features contribute to class 1 or 0. Anyone have any idea?", "votes": 6 }, { "id": 1503605, "postDate": "2021-09-05T14:17:12.377Z", "content": "0 - biopsy showed that the gene is not methylated. 1 - biopsy showed that the gene is methylated
", "rawMarkdown": "0 - biopsy showed that the gene is not methylated. 1 - biopsy showed that the gene is methylated" } ] }, { "id": 1503228, "postDate": "2021-09-05T06:56:18.380Z", "content": "What are we looking for then, in the images ?
", "rawMarkdown": "What are we looking for then, in the images ?" }, { "id": 1470576, "postDate": "2021-08-13T15:44:08.560Z", "content": "Thanks for sharing the visualizations. I am still not able to distinguish between scans with and without MGMT promoter methylation. What should I be looking for in the MRI scans?
", "rawMarkdown": "Thanks for sharing the visualizations. I am still not able to distinguish between scans with and without MGMT promoter methylation. What should I be looking for in the MRI scans?", "replies": [ { "id": 1472907, "postDate": "2021-08-15T07:50:03.847Z", "content": "Hey @novice03 I think the updated discussion post might shed some light on this. ;D
", "rawMarkdown": "Hey @novice03 I think the updated discussion post might shed some light on this. ;D", "votes": 1 }, { "id": 1476449, "postDate": "2021-08-17T06:02:33.247Z", "content": "Really helpful, thanks for sharing. I finally understood the competition. 👍
", "rawMarkdown": "Really helpful, thanks for sharing. I finally understood the competition. 👍" } ] }, { "id": 1476471, "postDate": "2021-08-17T06:15:45.887Z", "rawMarkdown": "", "votes": 3, "isDeleted": true } ], "comments": [ { "id": 1481279, "author_name": "Darien Schettler", "author_url": "", "post_date": "2021-08-19T12:09:15.567000", "content": "My super (overly)simple understanding is that MGMT promoter methylation is strongly correlated with chemotherapy success.
\nTherefore.
\nWe are classifying whether a tumour is likely to respond to chemotherapy or not.
", "votes": 6, "replies": [ { "id": 1489087, "author_name": "Ayush Thakur", "author_url": "", "post_date": "2021-08-24T17:40:36.023000", "content": "Agreed. I guess by now most of us are certain about the problem statement.
", "votes": 0, "replies": [] } ] }, { "id": 1472631, "author_name": "David Roberts", "author_url": "", "post_date": "2021-08-15T02:50:31.837000", "content": "I've looked at hundreds of series in both sets and have only seen a couple studies that I could not immediately identify a tumor in.
\nI suspect the ones I didn't see tumors on were because they were either small, not imaged well, better identified on a different series .. or I just didn't recognize it. Damnit Jim, I'm a tech not a doctor!.
\nMy assumptions are:
\n\n\nOur task is to determine if there is some difference between glioblastomas with MGMT and those without.
\n
I totally agree with this and my quick research reflects that. I have updated the discussion post with more info and useful links. It might be useful.
", "votes": 0, "replies": [] }, { "id": 1503603, "author_name": "Zaakcii Ru", "author_url": "", "post_date": "2021-09-05T14:14:59.710000", "content": "As far as I understand, methylation is determined only by molecular analysis, for which you need to cut out a piece of the tumor. in those articles that I read, it is argued that at the moment no differences have been found between methylated and unmethylated tumors on MRI, because the process takes place at the DNA level. methylation - blocking the work of the mgmt gene,
\nwhich restores cancer cells during chemotherapy. Another option to find out that the gene is methylated is to look at MRI scans taken at different times - before and after the start of chemotherapy, which is impossible given that we do not have such data.
Thanks for this, Initially I also thought of some correlation between tumour and MGMT, but after looking at many images I concluded I might be wrong.
", "votes": 1, "replies": [ { "id": 1470615, "author_name": "novice03", "author_url": "", "post_date": "2021-08-13T16:11:30.840000", "content": "Just to confirm, there are tumors in all scans, and we are predicting whether there is MGMT promoter methylation, right?
", "votes": 0, "replies": [] }, { "id": 1470626, "author_name": "DeepUnderstanding", "author_url": "", "post_date": "2021-08-13T16:17:18.370000", "content": "Tumours in all scans? I don't think so, they are random I guess.
\nI am not sure though
I think so. I am still yet to understand the problem completely. See this: https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/252843#1441395
", "votes": 1, "replies": [] }, { "id": 1470640, "author_name": "DeepUnderstanding", "author_url": "", "post_date": "2021-08-13T16:26:48.793000", "content": "wow,thanks for this!
", "votes": 1, "replies": [] }, { "id": 1470702, "author_name": "Ayush Thakur", "author_url": "", "post_date": "2021-08-13T17:02:41.627000", "content": "I think we have a brain tumor in every scan. If you visit the table and just look through the scans, it's easy to find lesions/edema/tumorish stuff.
", "votes": 5, "replies": [] }, { "id": 1471559, "author_name": "Haw Keat", "author_url": "", "post_date": "2021-08-14T09:40:32.667000", "content": "My understanding is for those classify 0, you may or may not have tumour. For those with class 1, you definitely have tumour. The level of presence of MGMT is the factor deciding class 0 or 1. The question I am thinking now is what is the features contribute to class 1 or 0. Anyone have any idea?
", "votes": 6, "replies": [] }, { "id": 1503605, "author_name": "Zaakcii Ru", "author_url": "", "post_date": "2021-09-05T14:17:12.377000", "content": "0 - biopsy showed that the gene is not methylated. 1 - biopsy showed that the gene is methylated
", "votes": 0, "replies": [] } ] }, { "id": 1503228, "author_name": "Md Zarif Ul Alam", "author_url": "", "post_date": "2021-09-05T06:56:18.380000", "content": "What are we looking for then, in the images ?
", "votes": 0, "replies": [] }, { "id": 1470576, "author_name": "novice03", "author_url": "", "post_date": "2021-08-13T15:44:08.560000", "content": "Thanks for sharing the visualizations. I am still not able to distinguish between scans with and without MGMT promoter methylation. What should I be looking for in the MRI scans?
", "votes": 0, "replies": [ { "id": 1472907, "author_name": "Ayush Thakur", "author_url": "", "post_date": "2021-08-15T07:50:03.847000", "content": "Hey @novice03 I think the updated discussion post might shed some light on this. ;D
", "votes": 1, "replies": [] }, { "id": 1476449, "author_name": "novice03", "author_url": "", "post_date": "2021-08-17T06:02:33.247000", "content": "Really helpful, thanks for sharing. I finally understood the competition. 👍
", "votes": 0, "replies": [] } ] }, { "id": 1476471, "author_name": "", "author_url": "", "post_date": "2021-08-17T06:15:45.887000", "content": "", "votes": 3, "replies": [] } ], "raw_markdown_by_id": { "1470454": "My first impression about this competition was that we need to classify a given ID to have/not have a brain tumor. My intuition was that the presence of MGMT promoter methylation leads to a brain tumor and that MRI sequences without MGMT promoters do not have a brain tumor.\n\n## I was wrong. How?\n\nI had logged the entire dataset onto W&B Tables. You can find more information about this in this [discussion post](http://wandb.me/kaggle-256051). \n\n#### [Check out the logged Tables here ▶️](http://wandb.me/brain-tumor-tables)\n\n\n\nAfter spending numerous hours looking through the dataset, I got confused because I could see the tumor-like artifact in almost all sequences, irrespective of the MGMT value. \n\n## So what's MGMT?\n\nGlioblastoma is the most frequent malignant primary tumor in the brain. It has a very poor prognosis, with a median survival of less than a year. The current standard of care consists of surgical resection followed by radiotherapy in addition to **alkylating chemotherapy** with temozolomide.\n\n> 📍 Alkylating agents are compounds that work by adding an alkyl group to the guanine base of the DNA molecule, preventing the double helix strands from linking as they should. This causes breakage of the DNA strands, affecting the ability of the cancer cell to multiply.\n\nMGMT (O[6]-methylguanine-DNA methyltransferase) is a DNA repair enzyme. This enzyme rescues tumor cells from alkylating agent-induced damage, leading to chemotherapy resistance with alkylating agents. \n\n**Epigenetic silencing of the MGMT gene by promoter methylation** results in decreased MGMT protein expression which potentially increases therapy sensitivity.\n\n> 📍 Epigenetic silencing is a natural phenomenon in which the expression of genes is regulated through modifications of DNA, RNA.\n\n## Update: More on MGMT's connection with Glioblastoma \n\nThis is based on this [article](https://oncologypro.esmo.org/education-library/factsheets-on-biomarkers/mgmt-promoter-methylation-in-glioma).\n\n* \"The O6-methylguanine-DNA methyl-transferase (MGMT) gene is located on chromosome 10q26.3\" - obviously we can't see MGMT in the MRI sequences. \n\n* \"Importantly, epigenetic modification could silence the MGMT gene, commonly through methylation of the cytosine-phosphate-guanine (CpG) island at specific CpG sites within the MGMT gene promoter.\": Epigenetic modification (silencing) happens naturally. This methylation of MGMT gene promoter induces loss/low levels of functional MGMT protein due to which alkylating chemotherapy works on the patient.\n\n* \"Methylation of the MGMT gene promoter has been observed in approximately 50% of grade IV gliomas, commonly referred to as glioblastoma multiforme\": This somewhat feels consistent with the distribution of the labels in the training set. \n\n* \"Originally, the first evidence that associated MGMT methylation status and response to TMZ emerged from the ‘Stupp trial’ \": Here's the [paper on Stupp trial](https://pubmed.ncbi.nlm.nih.gov/15758010/). From the conclusion section of this paper: \" Patients with glioblastoma containing a methylated MGMT promoter benefited from temozolomide, whereas those who did not have a methylated MGMT promoter did not have such a benefit. \"\n\n\n([Source](https://oncologypro.esmo.org/var/esmo/storage/images/media/images-op/biomarkers/mgmt-promoter-methylation-in-glioma/long-term-survival-results-according-to-type-of-treatment-and-mgmt-status-in-the-stupp-trial/3328288-1-eng-GB/Long-term-survival-results-according-to-type-of-treatment-and-MGMT-status-in-the-Stupp-trial_i1200.jpg))\n\n## MGMT Testing Methods\n\n* \"To date, no large consensus exists on the optimal method of assessment of MGMT gene promoter methylation, and different assays are available for this purpose. Regardless of the type of the chosen technique, the preliminary phases consist of genomic DNA extraction, it's qualitative/quantitative check, and bisulfite conversion.\": This process must take time and involve surgery to extract DNA. If an MRI sequence can help detect MGMT that would be great. \n\n## MRI and MGMT Connection\n\nFrom the conclusion section of, [\"Open Access\nClinically Relevant Imaging Features for MGMT Promoter Methylation in Multiple Glioblastoma Studies: A Systematic Review and Meta-Analysis\"](http://www.ajnr.org/content/early/2018/07/12/ajnr.A5711)\n\n> MGMT promotor methylated glioblastoma is likely to show less aggressive imaging features than MGMT promotor unmethylated glioblastoma.\n\nI think the paper titled, [\"Structural and advanced imaging in predicting MGMT promoter methylation of primary glioblastoma: a region of interest based analysis](https://bmccancer.biomedcentral.com/articles/10.1186/s12885-018-4114-2) finds the correlation between image feature and presence of MGMT. \n\nI didn't get into the details of the paper, but from a quick scan I think that there are some minute differences in the tumor with and without MGMT.\n\n\n\n> My thoughts: Given deep learning specializes in extracting \"meaningful\" features, let's leave it to the models to do the job and focus on using good ML practices. Domain knowledge will be super helpful in pre-/post-processing.\n\nI hope it might help clear some confusion. ", "1481279": "My super (overly)simple understanding is that MGMT promoter methylation is strongly correlated with chemotherapy success.\n\nTherefore.\n\n**We are classifying whether a tumour is likely to respond to chemotherapy or not.**", "1472631": "I've looked at hundreds of series in both sets and have only seen a couple studies that I could not immediately identify a tumor in. \n\nI suspect the ones I didn't see tumors on were because they were either small, not imaged well, better identified on a different series .. or I just didn't recognize it. Damnit Jim, I'm a tech not a doctor!.\n\nMy ***assumptions*** are:\n- Since BRATS is a brain tumor study and the previous comp was about identifying tumor types .. that this comp likely has studies with tumors only.\n- The tumor types have been identified as glioblastoma (and maybe glioma) by radiologists and confirmed by biopsy or lab test.\n- All tumors in this dataset are glioma/glioblastoma.\n- Our task is to determine if there is some difference between glioblastomas with MGMT and those without.\n- The differences might lie in the tumor, around the tumor or elsewhere in the brain.", "1470599": "Thanks for this, Initially I also thought of some correlation between tumour and MGMT, but after looking at many images I concluded I might be wrong.", "1503228": "What are we looking for then, in the images ?", "1470576": "Thanks for sharing the visualizations. I am still not able to distinguish between scans with and without MGMT promoter methylation. What should I be looking for in the MRI scans?", "1476471": "" } }