train_labels.csv => contains 585 rows
\nSo, I am expecting the total training images should be 585
\nand corresponding to each image, the output is given, whether it has a tumor or not (OR MGMT value)
But, when we checked the Training data - it has 585 folders
\nThat is okay, which matches the train_labels.csv file.
those folders further contain sub-folders (4 sub-folders) - what are those and why?
\nThese look type of MRI images
\nand why they contain so many images, which one to use?
I thought, 585 folders will contain just 1 image (corresponding to labels).
\nBut this is not the case here.
Can anyone help me to understand the structure of the data and why so many images (when we have just 585 labels).
", "messages": [ { "id": 1387867, "postDate": "2021-07-14T13:37:04.287Z", "content": "train_labels.csv => contains 585 rows
\nSo, I am expecting the total training images should be 585
\nand corresponding to each image, the output is given, whether it has a tumor or not (OR MGMT value)
But, when we checked the Training data - it has 585 folders
\nThat is okay, which matches the train_labels.csv file.
those folders further contain sub-folders (4 sub-folders) - what are those and why?
\nThese look type of MRI images
\nand why they contain so many images, which one to use?
I thought, 585 folders will contain just 1 image (corresponding to labels).
\nBut this is not the case here.
Can anyone help me to understand the structure of the data and why so many images (when we have just 585 labels).
", "rawMarkdown": "train_labels.csv => contains 585 rows \nSo, I am expecting the total training images should be 585\nand corresponding to each image, the output is given, whether it has a tumor or not (OR MGMT value)\n\nBut, when we checked the Training data - it has 585 folders\nThat is okay, which matches the train_labels.csv file.\n\nthose folders further contain sub-folders (4 sub-folders) - what are those and why?\nThese look type of MRI images\nand why they contain so many images, which one to use?\n\nI thought, 585 folders will contain just 1 image (corresponding to labels).\nBut this is not the case here.\n\nCan anyone help me to understand the structure of the data and why so many images (when we have just 585 labels).", "votes": 21 }, { "id": 1388006, "postDate": "2021-07-14T15:26:06.077Z", "content": "As David and Rupesh have mentioned, each label corresponds to a study (which is a single patient in this case). And each folder relates to a different MRI series.
\nThe MRI series' in this competition are
\nThe differences here correspond to changes in the way the MRI machine captures images. It's too much detail to go into here so here is a great youtube video on the topic.
\nTo put it briefly, in T2 images water is bright, and in T1 images fat is bright. In FLAIR, cerebrospinal fluid is dark but the rest of the image looks like T2.
", "rawMarkdown": "As David and Rupesh have mentioned, each label corresponds to a study (which is a single patient in this case). And each folder relates to a different MRI series.\n\nThe MRI series' in this competition are \n\n- Fluid Attenuated Inversion Recovery (FLAIR)\n- T1-weighted pre-contrast\n- T1-weighted post-contrast\n- T2-weighted\n\nThe differences here correspond to changes in the way the MRI machine captures images. It's too much detail to go into here [so here is a great youtube video on the topic](https://www.youtube.com/watch?v=djAxjtN_7VE). \n\nTo put it briefly, in T2 images water is bright, and in T1 images fat is bright. In FLAIR, cerebrospinal fluid is dark but the rest of the image looks like T2.", "votes": 14, "replies": [ { "id": 1388510, "postDate": "2021-07-15T03:01:01.207Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 1388739, "postDate": "2021-07-15T07:29:16.790Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1389154, "postDate": "2021-07-15T13:36:14.780Z", "content": "Some MR sequences are acquired at the same time as others. The patient is not normally moved at all during an MR brain scan. So, you can expect that they are in the same place for all the sequences. If the patient does move, the technologist will re-align the magnet to fit key anatomical points and keep the images standardized. It's not like in CT where you have to align the x-ray beam and the patient perfectly to acquire slices in the correct plane. MR scans are all Multiplanar Reconstructions which can be 'fitted' to any angle via software.
", "rawMarkdown": "Some MR sequences are acquired at the same time as others. The patient is not normally moved at all during an MR brain scan. So, you can expect that they are in the same place for all the sequences. If the patient does move, the technologist will re-align the magnet to fit key anatomical points and keep the images standardized. It's not like in CT where you have to align the x-ray beam and the patient perfectly to acquire slices in the correct plane. MR scans are all Multiplanar Reconstructions which can be 'fitted' to any angle via software. ", "votes": 2 }, { "id": 1389652, "postDate": "2021-07-15T22:59:47.650Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1389677, "postDate": "2021-07-15T23:59:29.537Z", "content": "@weka511, you weren't wrong. There are some post contrast series. Generally, the ones with CE (Contrast Enhancement) in the title are the ones done after the administration of a contrast agent.
", "rawMarkdown": "@weka511, you weren't wrong. There are some post contrast series. Generally, the ones with CE (Contrast Enhancement) in the title are the ones done after the administration of a contrast agent." }, { "id": 1389737, "postDate": "2021-07-16T03:10:15.237Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 1391690, "postDate": "2021-07-18T00:15:51.813Z", "content": "For brain MRI studies, it is typical to inject the contrast and image relatively immediately (within seconds to a minute or so).
\nThere are some studies where you wait longer.
", "rawMarkdown": "For brain MRI studies, it is typical to inject the contrast and image relatively immediately (within seconds to a minute or so).\n\nThere are some studies where you wait longer.", "votes": 1 } ] } ] }, { "id": 1387906, "postDate": "2021-07-14T14:02:12.007Z", "content": "Each label corresponds to a single study. Each study has four series, each series has multiple images.
\nMR studies are collections of 'stacks' of images similar to CT. Each series is acquired or processed differently. Some series are planar reconstructions (sagittal, coronal or axial), which display the same tissue in different planes.
\nMore about anatomical planes here -> https://en.wikipedia.org/wiki/Anatomical_plane
\nOne of the series in this comp is 'Post-Contrast' .. meaning it was acquired after the administration of gadolinium (or other contrast agent). This is helpful to visualize vascularity.
\nSome of the images are blank, or don't contain enough anatomy to be useful. The task here will be to determine which images offer some diagnostic information.
", "rawMarkdown": "Each label corresponds to a single study. Each study has four series, each series has multiple images.\n\nMR studies are collections of 'stacks' of images similar to CT. Each series is acquired or processed differently. Some series are planar reconstructions (sagittal, coronal or axial), which display the same tissue in different planes. \n\nMore about anatomical planes here -> https://en.wikipedia.org/wiki/Anatomical_plane\n\nOne of the series in this comp is 'Post-Contrast' .. meaning it was acquired after the administration of gadolinium (or other contrast agent). This is helpful to visualize vascularity.\n\nSome of the images are blank, or don't contain enough anatomy to be useful. The task here will be to determine which images offer *some* diagnostic information.", "votes": 14, "replies": [ { "id": 1387949, "postDate": "2021-07-14T14:40:37.430Z", "content": "This was really helpful to read. Thank you for sharing this basic explanation!
", "rawMarkdown": "This was really helpful to read. Thank you for sharing this basic explanation!", "votes": 1 } ] }, { "id": 1387883, "postDate": "2021-07-14T13:48:09.060Z", "content": "If I understand it correctly, the main folders with five-digit titles are the patients, the four sub-folders are the types of MRI performed on them. The multiple images in the same sub-folder are just multiple scans of same MRI on same patient, so you're free to use any one (or all). There's also a lot of other data in DICOM files other than images that you might find useful to extract. I made a notebook in a previous competition that also dealt with DICOM files. You might find it helpful.
\nhttps://www.kaggle.com/rude009/working-with-dicom-data
", "rawMarkdown": "If I understand it correctly, the main folders with five-digit titles are the patients, the four sub-folders are the types of MRI performed on them. The multiple images in the same sub-folder are just multiple scans of same MRI on same patient, so you're free to use any one (or all). There's also a lot of other data in DICOM files other than images that you might find useful to extract. I made a notebook in a previous competition that also dealt with DICOM files. You might find it helpful.\n\nhttps://www.kaggle.com/rude009/working-with-dicom-data", "votes": 9, "replies": [ { "id": 1387957, "postDate": "2021-07-14T14:43:39.037Z", "content": "Thanks for mentioning the notebook. After taking a look at it, I definitely have a better understanding of DICOM files. It was easy to read. Thnks!
", "rawMarkdown": "Thanks for mentioning the notebook. After taking a look at it, I definitely have a better understanding of DICOM files. It was easy to read. Thnks!", "votes": 1 }, { "id": 1388704, "postDate": "2021-07-15T06:40:54.807Z", "content": "Thank you!!! This was really helpful :)
", "rawMarkdown": "Thank you!!! This was really helpful :)", "votes": 1 } ] } ], "comments": [ { "id": 1388006, "author_name": "Reuben Schmidt", "author_url": "", "post_date": "2021-07-14T15:26:06.077000", "content": "As David and Rupesh have mentioned, each label corresponds to a study (which is a single patient in this case). And each folder relates to a different MRI series.
\nThe MRI series' in this competition are
\nThe differences here correspond to changes in the way the MRI machine captures images. It's too much detail to go into here so here is a great youtube video on the topic.
\nTo put it briefly, in T2 images water is bright, and in T1 images fat is bright. In FLAIR, cerebrospinal fluid is dark but the rest of the image looks like T2.
", "votes": 14, "replies": [ { "id": 1388510, "author_name": "", "author_url": "", "post_date": "2021-07-15T03:01:01.207000", "content": "", "votes": 1, "replies": [] }, { "id": 1388739, "author_name": "", "author_url": "", "post_date": "2021-07-15T07:29:16.790000", "content": "", "votes": 0, "replies": [] }, { "id": 1389154, "author_name": "David Roberts", "author_url": "", "post_date": "2021-07-15T13:36:14.780000", "content": "Some MR sequences are acquired at the same time as others. The patient is not normally moved at all during an MR brain scan. So, you can expect that they are in the same place for all the sequences. If the patient does move, the technologist will re-align the magnet to fit key anatomical points and keep the images standardized. It's not like in CT where you have to align the x-ray beam and the patient perfectly to acquire slices in the correct plane. MR scans are all Multiplanar Reconstructions which can be 'fitted' to any angle via software.
", "votes": 2, "replies": [] }, { "id": 1389652, "author_name": "", "author_url": "", "post_date": "2021-07-15T22:59:47.650000", "content": "", "votes": 0, "replies": [] }, { "id": 1389677, "author_name": "David Roberts", "author_url": "", "post_date": "2021-07-15T23:59:29.537000", "content": "@weka511, you weren't wrong. There are some post contrast series. Generally, the ones with CE (Contrast Enhancement) in the title are the ones done after the administration of a contrast agent.
", "votes": 0, "replies": [] }, { "id": 1389737, "author_name": "", "author_url": "", "post_date": "2021-07-16T03:10:15.237000", "content": "", "votes": 0, "replies": [ { "id": 1391690, "author_name": "quadcore/Richard Epstein", "author_url": "", "post_date": "2021-07-18T00:15:51.813000", "content": "For brain MRI studies, it is typical to inject the contrast and image relatively immediately (within seconds to a minute or so).
\nThere are some studies where you wait longer.
", "votes": 1, "replies": [] } ] } ] }, { "id": 1387906, "author_name": "David Roberts", "author_url": "", "post_date": "2021-07-14T14:02:12.007000", "content": "Each label corresponds to a single study. Each study has four series, each series has multiple images.
\nMR studies are collections of 'stacks' of images similar to CT. Each series is acquired or processed differently. Some series are planar reconstructions (sagittal, coronal or axial), which display the same tissue in different planes.
\nMore about anatomical planes here -> https://en.wikipedia.org/wiki/Anatomical_plane
\nOne of the series in this comp is 'Post-Contrast' .. meaning it was acquired after the administration of gadolinium (or other contrast agent). This is helpful to visualize vascularity.
\nSome of the images are blank, or don't contain enough anatomy to be useful. The task here will be to determine which images offer some diagnostic information.
", "votes": 14, "replies": [ { "id": 1387949, "author_name": "ElenaEB", "author_url": "", "post_date": "2021-07-14T14:40:37.430000", "content": "This was really helpful to read. Thank you for sharing this basic explanation!
", "votes": 1, "replies": [] } ] }, { "id": 1387883, "author_name": "Rupesh Deshmukh", "author_url": "", "post_date": "2021-07-14T13:48:09.060000", "content": "If I understand it correctly, the main folders with five-digit titles are the patients, the four sub-folders are the types of MRI performed on them. The multiple images in the same sub-folder are just multiple scans of same MRI on same patient, so you're free to use any one (or all). There's also a lot of other data in DICOM files other than images that you might find useful to extract. I made a notebook in a previous competition that also dealt with DICOM files. You might find it helpful.
\nhttps://www.kaggle.com/rude009/working-with-dicom-data
", "votes": 9, "replies": [ { "id": 1387957, "author_name": "ElenaEB", "author_url": "", "post_date": "2021-07-14T14:43:39.037000", "content": "Thanks for mentioning the notebook. After taking a look at it, I definitely have a better understanding of DICOM files. It was easy to read. Thnks!
", "votes": 1, "replies": [] }, { "id": 1388704, "author_name": "Pranav Kushare", "author_url": "", "post_date": "2021-07-15T06:40:54.807000", "content": "Thank you!!! This was really helpful :)
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "1387867": "train_labels.csv => contains 585 rows \nSo, I am expecting the total training images should be 585\nand corresponding to each image, the output is given, whether it has a tumor or not (OR MGMT value)\n\nBut, when we checked the Training data - it has 585 folders\nThat is okay, which matches the train_labels.csv file.\n\nthose folders further contain sub-folders (4 sub-folders) - what are those and why?\nThese look type of MRI images\nand why they contain so many images, which one to use?\n\nI thought, 585 folders will contain just 1 image (corresponding to labels).\nBut this is not the case here.\n\nCan anyone help me to understand the structure of the data and why so many images (when we have just 585 labels).", "1388006": "As David and Rupesh have mentioned, each label corresponds to a study (which is a single patient in this case). And each folder relates to a different MRI series.\n\nThe MRI series' in this competition are \n\n- Fluid Attenuated Inversion Recovery (FLAIR)\n- T1-weighted pre-contrast\n- T1-weighted post-contrast\n- T2-weighted\n\nThe differences here correspond to changes in the way the MRI machine captures images. It's too much detail to go into here [so here is a great youtube video on the topic](https://www.youtube.com/watch?v=djAxjtN_7VE). \n\nTo put it briefly, in T2 images water is bright, and in T1 images fat is bright. In FLAIR, cerebrospinal fluid is dark but the rest of the image looks like T2.", "1387906": "Each label corresponds to a single study. Each study has four series, each series has multiple images.\n\nMR studies are collections of 'stacks' of images similar to CT. Each series is acquired or processed differently. Some series are planar reconstructions (sagittal, coronal or axial), which display the same tissue in different planes. \n\nMore about anatomical planes here -> https://en.wikipedia.org/wiki/Anatomical_plane\n\nOne of the series in this comp is 'Post-Contrast' .. meaning it was acquired after the administration of gadolinium (or other contrast agent). This is helpful to visualize vascularity.\n\nSome of the images are blank, or don't contain enough anatomy to be useful. The task here will be to determine which images offer *some* diagnostic information.", "1387883": "If I understand it correctly, the main folders with five-digit titles are the patients, the four sub-folders are the types of MRI performed on them. The multiple images in the same sub-folder are just multiple scans of same MRI on same patient, so you're free to use any one (or all). There's also a lot of other data in DICOM files other than images that you might find useful to extract. I made a notebook in a previous competition that also dealt with DICOM files. You might find it helpful.\n\nhttps://www.kaggle.com/rude009/working-with-dicom-data" } }