![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2593251%2F37f4f58c3d12bddc10cf2c54abbcf8c2%2Ft1ce_aca.png?generation=1757212978827357&alt=media\")
Based on my understanding, the current challenge is designed with input that can be any of the four modalities: T1, T2, MRA and CTA, and the model is expected to work independently on a single modality input. However, this is argubly against clinical guidelines as no clinicians will make aneurysm diagnosis sorely on T1 or T2 modalities.
\nMay I know is our model expected to work on single modality of T1 and T2 to detect aneurysm? And if so, why are we not allowed to use the clinically relevant modalities of MRA and CTA for a patient and have to rely on T1 or T2?
\nThanks!
", "messages": [ { "id": 3281091, "postDate": "2025-09-03T16:05:23.050Z", "content": "Based on my understanding, the current challenge is designed with input that can be any of the four modalities: T1, T2, MRA and CTA, and the model is expected to work independently on a single modality input. However, this is argubly against clinical guidelines as no clinicians will make aneurysm diagnosis sorely on T1 or T2 modalities.
\nMay I know is our model expected to work on single modality of T1 and T2 to detect aneurysm? And if so, why are we not allowed to use the clinically relevant modalities of MRA and CTA for a patient and have to rely on T1 or T2?
\nThanks!
", "rawMarkdown": "Based on my understanding, the current challenge is designed with input that can be any of the four modalities: T1, T2, MRA and CTA, and the model is expected to work independently on a single modality input. However, this is argubly against clinical guidelines as no clinicians will make aneurysm diagnosis sorely on T1 or T2 modalities.\n\nMay I know is our model expected to work on single modality of T1 and T2 to detect aneurysm? And if so, why are we not allowed to use the clinically relevant modalities of MRA and CTA for a patient and have to rely on T1 or T2?\n\nThanks!", "votes": 10 }, { "id": 3281326, "postDate": "2025-09-04T03:56:21.160Z", "content": "While T2 and T1post contrast are not the most optimal modalities to detect aneurysms, they can be seen on these sequences fairly well in most cases and many radiologists can make detect them on these sequences . We wanted to increase the potential for algorithms created by this challenge to be used as opportunistic screening since patients getting brain MRIs for other reasons don't usually get a CTA/MRA.
\nTake a look at this pinned post regarding clinical overview of the problem - https://www.kaggle.com/competitions/rsna-intracranial-aneurysm-detection/discussion/591648#3280608
", "rawMarkdown": "While T2 and T1post contrast are not the most optimal modalities to detect aneurysms, they can be seen on these sequences fairly well in most cases and many radiologists can make detect them on these sequences . We wanted to increase the potential for algorithms created by this challenge to be used as opportunistic screening since patients getting brain MRIs for other reasons don't usually get a CTA/MRA. \n\nTake a look at this pinned post regarding clinical overview of the problem - https://www.kaggle.com/competitions/rsna-intracranial-aneurysm-detection/discussion/591648#3280608", "votes": 1, "replies": [ { "id": 3281440, "postDate": "2025-09-04T09:52:47.770Z", "content": "Hi Jeff, Thank you for your clarification. But in the context of your challenge, the vessels are not clearly visible in T1 modality, then how can the vessel location of an aneurysm be classified? The anterior vessels themselves are barely distinguishable, let alone determining the left or right vessel. Similarly, the communication arteries (Pcom and Acom) are also not clearly visible, and it is beyond the modality to determine whether the vessel location is Pcom or ICA.
\nAnd even if certain aneurysms may be detectable in T1 or T2, those need to be special types of aneurysm in terms of location and size. Are only specific types of aneurysms labeled in T1 or T2 and are they modality-specific? (or are the labels determined via standard modalities like MRA and CTA of that patient)
\nThank you!
", "rawMarkdown": "Hi Jeff, Thank you for your clarification. But in the context of your challenge, the vessels are not clearly visible in T1 modality, then how can the vessel location of an aneurysm be classified? The anterior vessels themselves are barely distinguishable, let alone determining the left or right vessel. Similarly, the communication arteries (Pcom and Acom) are also not clearly visible, and it is beyond the modality to determine whether the vessel location is Pcom or ICA.\n\nAnd even if certain aneurysms may be detectable in T1 or T2, those need to be special types of aneurysm in terms of location and size. Are only specific types of aneurysms labeled in T1 or T2 and are they modality-specific? (or are the labels determined via standard modalities like MRA and CTA of that patient)\n\nThank you!", "votes": 3, "replies": [ { "id": 3281588, "postDate": "2025-09-04T14:37:55.073Z", "content": "Without having medical expertise, I would wager that the brain tissue also experiences change, even if only minute.
\nI suspect that, if such changes actually occur, this would not be visible to trained human eyes - but to machine learning models, this might be more than enough.
\nAgain, I am not medically educated and can't say whether or not brain tissue does alter from an aneurysm (burst or in-tact), but I believe it is a hypothesis worth testing regardless.
Just to add a clinical perspective here: You’re right that on T2WI the slice thickness is relatively large, and in some cases it can indeed be difficult to distinguish, for example, between the left and right ACAs. That said, aneurysms above a certain size often show a flow void, so some can actually be seen quite easily. Having a mix of challenging targets (sometimes nearly impossible) and easier targets feels fairly typical for many deep learning tasks.
\nIn addition, as shown in the attached image, T1CE provides good visualization, and with the vascular contrast enhancement, the vessels can be seen much more clearly.
\n
Glad to see you on board with this competition - wishing you best of luck! I'm certainly having more issues with this one compared to SPR.
", "rawMarkdown": "Glad to see you on board with this competition - wishing you best of luck! I'm certainly having more issues with this one compared to SPR.", "votes": 1 }, { "id": 3283448, "postDate": "2025-09-07T16:53:42.870Z", "content": "Hi, thank you for the sharing. My main concern is are the aneurysms labeled in T1 and T2 modality-specific. If so, then the aneurysms are limited to a special subset of intracranial aneurysms in terms of location and size, as you have pointed out. Or are they labeled/diagnosed using the reference modality (MRA or CTA) and then generated the label for T1 and T2? This needs to be clarified by the organizers.
", "rawMarkdown": "Hi, thank you for the sharing. My main concern is are the aneurysms labeled in T1 and T2 modality-specific. If so, then the aneurysms are limited to a special subset of intracranial aneurysms in terms of location and size, as you have pointed out. Or are they labeled/diagnosed using the reference modality (MRA or CTA) and then generated the label for T1 and T2? This needs to be clarified by the organizers.", "votes": 1 }, { "id": 3283852, "postDate": "2025-09-08T01:15:02.707Z", "content": "Reference CTA or MRA were available when reviewing T2/T1post sequences. Aneurysms were only marked as present on T1/T1post sequences if they were also visually confirmed by a radiologist on those images.
", "rawMarkdown": "Reference CTA or MRA were available when reviewing T2/T1post sequences. Aneurysms were only marked as present on T1/T1post sequences if they were also visually confirmed by a radiologist on those images. ", "votes": 4 } ] } ] } ], "comments": [ { "id": 3281326, "author_name": "JeffRudie", "author_url": "", "post_date": "2025-09-04T03:56:21.160000", "content": "While T2 and T1post contrast are not the most optimal modalities to detect aneurysms, they can be seen on these sequences fairly well in most cases and many radiologists can make detect them on these sequences . We wanted to increase the potential for algorithms created by this challenge to be used as opportunistic screening since patients getting brain MRIs for other reasons don't usually get a CTA/MRA.
\nTake a look at this pinned post regarding clinical overview of the problem - https://www.kaggle.com/competitions/rsna-intracranial-aneurysm-detection/discussion/591648#3280608
", "votes": 1, "replies": [ { "id": 3281440, "author_name": "MakeLoveAndPeace", "author_url": "", "post_date": "2025-09-04T09:52:47.770000", "content": "Hi Jeff, Thank you for your clarification. But in the context of your challenge, the vessels are not clearly visible in T1 modality, then how can the vessel location of an aneurysm be classified? The anterior vessels themselves are barely distinguishable, let alone determining the left or right vessel. Similarly, the communication arteries (Pcom and Acom) are also not clearly visible, and it is beyond the modality to determine whether the vessel location is Pcom or ICA.
\nAnd even if certain aneurysms may be detectable in T1 or T2, those need to be special types of aneurysm in terms of location and size. Are only specific types of aneurysms labeled in T1 or T2 and are they modality-specific? (or are the labels determined via standard modalities like MRA and CTA of that patient)
\nThank you!
", "votes": 3, "replies": [ { "id": 3281588, "author_name": "Skelp", "author_url": "", "post_date": "2025-09-04T14:37:55.073000", "content": "Without having medical expertise, I would wager that the brain tissue also experiences change, even if only minute.
\nI suspect that, if such changes actually occur, this would not be visible to trained human eyes - but to machine learning models, this might be more than enough.
\nAgain, I am not medically educated and can't say whether or not brain tissue does alter from an aneurysm (burst or in-tact), but I believe it is a hypothesis worth testing regardless.
Just to add a clinical perspective here: You’re right that on T2WI the slice thickness is relatively large, and in some cases it can indeed be difficult to distinguish, for example, between the left and right ACAs. That said, aneurysms above a certain size often show a flow void, so some can actually be seen quite easily. Having a mix of challenging targets (sometimes nearly impossible) and easier targets feels fairly typical for many deep learning tasks.
\nIn addition, as shown in the attached image, T1CE provides good visualization, and with the vascular contrast enhancement, the vessels can be seen much more clearly.
\n
Glad to see you on board with this competition - wishing you best of luck! I'm certainly having more issues with this one compared to SPR.
", "votes": 1, "replies": [] }, { "id": 3283448, "author_name": "MakeLoveAndPeace", "author_url": "", "post_date": "2025-09-07T16:53:42.870000", "content": "Hi, thank you for the sharing. My main concern is are the aneurysms labeled in T1 and T2 modality-specific. If so, then the aneurysms are limited to a special subset of intracranial aneurysms in terms of location and size, as you have pointed out. Or are they labeled/diagnosed using the reference modality (MRA or CTA) and then generated the label for T1 and T2? This needs to be clarified by the organizers.
", "votes": 1, "replies": [] }, { "id": 3283852, "author_name": "JeffRudie", "author_url": "", "post_date": "2025-09-08T01:15:02.707000", "content": "Reference CTA or MRA were available when reviewing T2/T1post sequences. Aneurysms were only marked as present on T1/T1post sequences if they were also visually confirmed by a radiologist on those images.
", "votes": 4, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "3281091": "Based on my understanding, the current challenge is designed with input that can be any of the four modalities: T1, T2, MRA and CTA, and the model is expected to work independently on a single modality input. However, this is argubly against clinical guidelines as no clinicians will make aneurysm diagnosis sorely on T1 or T2 modalities.\n\nMay I know is our model expected to work on single modality of T1 and T2 to detect aneurysm? And if so, why are we not allowed to use the clinically relevant modalities of MRA and CTA for a patient and have to rely on T1 or T2?\n\nThanks!", "3281326": "While T2 and T1post contrast are not the most optimal modalities to detect aneurysms, they can be seen on these sequences fairly well in most cases and many radiologists can make detect them on these sequences . We wanted to increase the potential for algorithms created by this challenge to be used as opportunistic screening since patients getting brain MRIs for other reasons don't usually get a CTA/MRA. \n\nTake a look at this pinned post regarding clinical overview of the problem - https://www.kaggle.com/competitions/rsna-intracranial-aneurysm-detection/discussion/591648#3280608" } }