I have spent the last couple of days reviewing my preprocessing of the MR series and found a certain number of issues similar to those I raised for CT series in this topic.
\nI have attached below a list of 20 aneurysm locations from modality MR series where the coordinates are wrong, the labels might be incorrect (left/right) or the aneurysms have been allocated to the wrong slice.
Series 1.2.826.0.1.3680043.8.498.75294325392457179365040684378207706807: all the DICOM files when opened and extracted are black with parts of a diagonal white line (a bit like a shooting star). The number of bits stored is 16, so this is not a bit shift problem. While there are some series that are unusable for various quality issues, it would be a pity to lose this one, as it contains two aneurysms. Could it be possible to re-produce/fix the series as it was initially intended?
Series 1.2.826.0.1.3680043.8.498.72399220582146674554688600955035038848: this seems to be the only series in the whole sample (apart from the multiframe DICOMs I reported here) where the ImagePositionPatient data cannot be used to order the slices correctly (it starts at the middle of the brain for two slices, then moves to the neck for two slices, etc.). The only way to order them correctly would be to use the ascending InstanceNumbers. This method cannot be used for the whole sample, as some series are ordered in ascending InstanceNumbers, some in descending InstanceNumbers, while others do not follow any of these two patterns. Could it be possible to correct the ImagePositionPatient vector for this series so that it can be ordered in the same manner as the rest of the sample (this series contains one aneurysm).
Could you please address these issues at your earliest convenience.
", "messages": [ { "id": 3269864, "postDate": "2025-08-15T07:58:24.030Z", "content": "I have spent the last couple of days reviewing my preprocessing of the MR series and found a certain number of issues similar to those I raised for CT series in this topic.
\nI have attached below a list of 20 aneurysm locations from modality MR series where the coordinates are wrong, the labels might be incorrect (left/right) or the aneurysms have been allocated to the wrong slice.
Series 1.2.826.0.1.3680043.8.498.75294325392457179365040684378207706807: all the DICOM files when opened and extracted are black with parts of a diagonal white line (a bit like a shooting star). The number of bits stored is 16, so this is not a bit shift problem. While there are some series that are unusable for various quality issues, it would be a pity to lose this one, as it contains two aneurysms. Could it be possible to re-produce/fix the series as it was initially intended?
Series 1.2.826.0.1.3680043.8.498.72399220582146674554688600955035038848: this seems to be the only series in the whole sample (apart from the multiframe DICOMs I reported here) where the ImagePositionPatient data cannot be used to order the slices correctly (it starts at the middle of the brain for two slices, then moves to the neck for two slices, etc.). The only way to order them correctly would be to use the ascending InstanceNumbers. This method cannot be used for the whole sample, as some series are ordered in ascending InstanceNumbers, some in descending InstanceNumbers, while others do not follow any of these two patterns. Could it be possible to correct the ImagePositionPatient vector for this series so that it can be ordered in the same manner as the rest of the sample (this series contains one aneurysm).
Could you please address these issues at your earliest convenience.
", "rawMarkdown": "I have spent the last couple of days reviewing my preprocessing of the MR series and found a certain number of issues similar to those I raised for CT series in [this topic](https://www.kaggle.com/competitions/rsna-intracranial-aneurysm-detection/discussion/596492).\n\n1. I have attached below a list of 20 aneurysm locations from modality MR series where the coordinates are wrong, the labels might be incorrect (left/right) or the aneurysms have been allocated to the wrong slice.\n\n2. Series 1.2.826.0.1.3680043.8.498.75294325392457179365040684378207706807: all the DICOM files when opened and extracted are black with parts of a diagonal white line (a bit like a shooting star). The number of bits stored is 16, so this is not a bit shift problem. While there are some series that are unusable for various quality issues, it would be a pity to lose this one, as it contains two aneurysms. Could it be possible to re-produce/fix the series as it was initially intended?\n\n3. Series 1.2.826.0.1.3680043.8.498.72399220582146674554688600955035038848: this seems to be the only series in the whole sample (apart from the multiframe DICOMs I reported [here](https://www.kaggle.com/competitions/rsna-intracranial-aneurysm-detection/discussion/598828)) where the ImagePositionPatient data cannot be used to order the slices correctly (it starts at the middle of the brain for two slices, then moves to the neck for two slices, etc.). The only way to order them correctly would be to use the ascending InstanceNumbers. This method cannot be used for the whole sample, as some series are ordered in ascending InstanceNumbers, some in descending InstanceNumbers, while others do not follow any of these two patterns. Could it be possible to correct the ImagePositionPatient vector for this series so that it can be ordered in the same manner as the rest of the sample (this series contains one aneurysm).\n\nCould you please address these issues at your earliest convenience.", "votes": 10 }, { "id": 3270015, "postDate": "2025-08-15T14:34:22.353Z", "content": "Thanks for the info. More stuff for us to look into!
", "rawMarkdown": "Thanks for the info. More stuff for us to look into!", "votes": 1 } ], "comments": [ { "id": 3270015, "author_name": "Evan Calabrese", "author_url": "", "post_date": "2025-08-15T14:34:22.353000", "content": "Thanks for the info. More stuff for us to look into!
", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "3269864": "I have spent the last couple of days reviewing my preprocessing of the MR series and found a certain number of issues similar to those I raised for CT series in [this topic](https://www.kaggle.com/competitions/rsna-intracranial-aneurysm-detection/discussion/596492).\n\n1. I have attached below a list of 20 aneurysm locations from modality MR series where the coordinates are wrong, the labels might be incorrect (left/right) or the aneurysms have been allocated to the wrong slice.\n\n2. Series 1.2.826.0.1.3680043.8.498.75294325392457179365040684378207706807: all the DICOM files when opened and extracted are black with parts of a diagonal white line (a bit like a shooting star). The number of bits stored is 16, so this is not a bit shift problem. While there are some series that are unusable for various quality issues, it would be a pity to lose this one, as it contains two aneurysms. Could it be possible to re-produce/fix the series as it was initially intended?\n\n3. Series 1.2.826.0.1.3680043.8.498.72399220582146674554688600955035038848: this seems to be the only series in the whole sample (apart from the multiframe DICOMs I reported [here](https://www.kaggle.com/competitions/rsna-intracranial-aneurysm-detection/discussion/598828)) where the ImagePositionPatient data cannot be used to order the slices correctly (it starts at the middle of the brain for two slices, then moves to the neck for two slices, etc.). The only way to order them correctly would be to use the ascending InstanceNumbers. This method cannot be used for the whole sample, as some series are ordered in ascending InstanceNumbers, some in descending InstanceNumbers, while others do not follow any of these two patterns. Could it be possible to correct the ImagePositionPatient vector for this series so that it can be ordered in the same manner as the rest of the sample (this series contains one aneurysm).\n\nCould you please address these issues at your earliest convenience.", "3270015": "Thanks for the info. More stuff for us to look into!" } }