Hello,
\nJust to confirm that I am understanding the problem correctly:
Am I understanding this correctly?
\nThanks!
", "messages": [ { "id": 1178626, "postDate": "2021-01-31T03:16:04.613Z", "content": "Hello,
\nJust to confirm that I am understanding the problem correctly:
Am I understanding this correctly?
\nThanks!
", "rawMarkdown": "Hello, \nJust to confirm that I am understanding the problem correctly: \n1. There are many cells in a single image (let's say 5 cells) \n2. Either all or subset of the cells are coded in green (let's say 3 cells)\n3. So, the label(s) (e.g., 1, 4, 5) tagged to this image only applies to the 3 green cells?\n\nAm I understanding this correctly? \n\nThanks!", "votes": 17 }, { "id": 1179526, "postDate": "2021-01-31T15:54:19.303Z", "content": "The labels are the organelles/structures in which the proteins are located. So you are correct in assuming that the labels only apply to the cells where green is present (as the green is the protein).
\nIf there are 4 cells and 3 have green staining and the image-level labels are Mitochondria, and Nucleoplasm.
\nCell 1:
\n - Green looks to be in the Mitochondria. Therefore, the cell level is Mitochondria.
Cell 2:
\n - Green looks to be in the Nucleoplasm. Therefore, the cell level is Nucleoplasm.
Cell 3:
\n - Green looks to be in the Nucleoplasm and Mitochondria. Therefore, the cell level label is Nucleoplasm and Mitochondria
Cell 4:
\n - No green or green is not present in any organelle. Therefore, the cell level label is Negative.
Thanks for the awesome feedback! Just a few more observations & questions that I still have are:
\nObservations:
\nSo,
\nA. If we have images with label 10, the green-channel will have overlap with the red-channel, both of which are \"microtubule\"
\nB. If we have images with label 6, the green-channel will have overlap with the yellow-channel, both of which are \"endoplasmic reticulum\"
\nC. 1) If an image has neither label 10 nor 6, but do have red and yellow channel highlighting portions of the image, we can use the regions highlighted by the red and yellow for label 10 and 6 respectively.
Is this correct?
", "rawMarkdown": "Thanks for the awesome feedback! Just a few more observations & questions that I still have are: \n\nObservations: \n1. Red channel is \"microtubule\", which corresponds to label 10 (a label that will only apply to the green channel).\n2. Yellow channel is \"endoplasmic reticulum\" which corresponds to label 6 (a label that will only apply to the green channel). \n\nSo, \nA. If we have images with label 10, the green-channel will have overlap with the red-channel, both of which are \"microtubule\"\nB. If we have images with label 6, the green-channel will have overlap with the yellow-channel, both of which are \"endoplasmic reticulum\"\nC. 1) If an image has neither label 10 nor 6, but do have red and yellow channel highlighting portions of the image, we can use the regions highlighted by the red and yellow for label 10 and 6 respectively. \n\nIs this correct? ", "votes": 1, "replies": [ { "id": 1184158, "postDate": "2021-02-03T12:18:15.620Z", "content": "A+B. You are correct, the red channel will help you to identify a microtubule pattern (label 10) in the green channel, and the yellow channel will help you identify an ER pattern (label 6) in the green channel. The reference channels (red/green/blue) can also help you to get an overview of the cell - certain patterns will always be distributed in a certain way in correlation to these. For example, labels 2 to 5 are all within the nucleus (i.e. the region of each cell with a signal in the blue channel).
\nC. All images have all the four channels, and signals from the markers (blue, yellow, red) are present in all cells in the image, independent of the green channel that you are classifying, in order to help you identify where the cells are, as well as where certain structures and regions within the cells are. This can in turn help you to segment the cells and to classify each cell to one or more label(s) according to the signal in the green channel. Note that each of the image-level labels can be present in all or in just a fraction of the individual cells in the image, and that some individual cells may also have additional labels, that are not mentioned among the image-level labels.
", "rawMarkdown": "A+B. You are correct, the red channel will help you to identify a microtubule pattern (label 10) in the green channel, and the yellow channel will help you identify an ER pattern (label 6) in the green channel. The reference channels (red/green/blue) can also help you to get an overview of the cell - certain patterns will always be distributed in a certain way in correlation to these. For example, labels 2 to 5 are all within the nucleus (i.e. the region of each cell with a signal in the blue channel).\n\nC. All images have all the four channels, and signals from the markers (blue, yellow, red) are present in all cells in the image, independent of the green channel that you are classifying, in order to help you identify where the cells are, as well as where certain structures and regions within the cells are. This can in turn help you to segment the cells and to classify each cell to one or more label(s) according to the signal in the green channel. Note that each of the image-level labels can be present in all or in just a fraction of the individual cells in the image, and that some individual cells may also have additional labels, that are not mentioned among the image-level labels.", "votes": 3 }, { "id": 1186834, "postDate": "2021-02-05T04:35:37.817Z", "content": "Hi @uaxelsson, can we ask a quick clarification on this part?:
\n\n\nsome individual cells may also have additional labels, that are not mentioned among the image-level labels.
\n
We understand that the additional label might be negative, but other than that it would mean that the image-level labels were not fully accurate and additional labels where added while tagging segmented cells, is that correct? Do we have an indication of how much the labels in the test set differ in this regard from the training set? @dschettler8845
", "rawMarkdown": "Hi @uaxelsson, can we ask a quick clarification on this part?:\n> some individual cells may also have additional labels, that are not mentioned among the image-level labels.\n\nWe understand that the additional label might be negative, but other than that it would mean that the image-level labels were not fully accurate and additional labels where added while tagging segmented cells, is that correct? Do we have an indication of how much the labels in the test set differ in this regard from the training set? @dschettler8845", "votes": 2 }, { "id": 1187705, "postDate": "2021-02-05T16:41:51.453Z", "content": "I would like to see the answer to this as well. Is this only w.r.t. to the Negative label?
\nThanks for tagging me Darek!
", "rawMarkdown": "I would like to see the answer to this as well. Is this only w.r.t. to the **Negative** label? \n\nThanks for tagging me Darek!" }, { "id": 1187717, "postDate": "2021-02-05T16:51:32.180Z", "content": "Yes there may be images where a single cell have a label that was neglected at the image level (for example a weak cytosolic pattern). Managing such weak-labels is part of this challenge, and it's impact is something we will analyze at the end of the challenge.
", "rawMarkdown": "Yes there may be images where a single cell have a label that was neglected at the image level (for example a weak cytosolic pattern). Managing such weak-labels is part of this challenge, and it's impact is something we will analyze at the end of the challenge.", "votes": 4 } ] }, { "id": 1179066, "postDate": "2021-01-31T09:39:22.900Z", "content": "From what I understand, some classes are more visible in other channels, e.g. Endoplasmic reticulum in yellow, see https://www.kaggle.com/lnhtrang/single-cell-patterns - other than that I think you are right that image-level classes can apply to a sub-set of the cells.
", "rawMarkdown": "From what I understand, some classes are more visible in other channels, e.g. Endoplasmic reticulum in yellow, see https://www.kaggle.com/lnhtrang/single-cell-patterns - other than that I think you are right that image-level classes can apply to a sub-set of the cells.", "votes": 2, "replies": [ { "id": 1182088, "postDate": "2021-02-02T09:46:52.597Z", "content": "Yes, but please note that it is always the pattern(s) in the green channel that you should classify. Yellow will show you what the Endoplasmic reticulum looks like in each cell. It is a marker present in all images, to help identify proteins of interest (in green) that also localize to the endoplasmic reticulum.
", "rawMarkdown": "Yes, but please note that it is always the pattern(s) in the green channel that you should classify. Yellow will show you what the Endoplasmic reticulum looks like in each cell. It is a marker present in all images, to help identify proteins of interest (in green) that also localize to the endoplasmic reticulum.", "votes": 3 }, { "id": 1207467, "postDate": "2021-02-17T21:03:26.723Z", "content": "Is it safe to assume that segmented cells without any signal from the green channel will have a negative(18) label with confidence 1? @UAxelsson
", "rawMarkdown": "Is it safe to assume that segmented cells without any signal from the green channel will have a negative(18) label with confidence 1? @UAxelsson" }, { "id": 1211560, "postDate": "2021-02-20T10:29:14.037Z", "content": "@uaxelsson does this mean that we should train models only on green images, when you say \"should classify\" ?
", "rawMarkdown": "@uaxelsson does this mean that we should train models only on green images, when you say \"should classify\" ?" }, { "id": 1213899, "postDate": "2021-02-22T12:26:19.610Z", "content": "@mohammedaminetriki The green channel is the channel for the protein of interest. What we want to know is what cellular compartments the protein is localizing to.
\nThe channels other than green are our reference points when annotating so they will probably help your model but you are in no way forced to use them.
Wow, that clarified a ton!
\nThanks for the detailed response @dschettler8845 and @lopuhin !!
The labels are the organelles/structures in which the proteins are located. So you are correct in assuming that the labels only apply to the cells where green is present (as the green is the protein).
\nIf there are 4 cells and 3 have green staining and the image-level labels are Mitochondria, and Nucleoplasm.
\nCell 1:
\n - Green looks to be in the Mitochondria. Therefore, the cell level is Mitochondria.
Cell 2:
\n - Green looks to be in the Nucleoplasm. Therefore, the cell level is Nucleoplasm.
Cell 3:
\n - Green looks to be in the Nucleoplasm and Mitochondria. Therefore, the cell level label is Nucleoplasm and Mitochondria
Cell 4:
\n - No green or green is not present in any organelle. Therefore, the cell level label is Negative.
Thanks for the awesome feedback! Just a few more observations & questions that I still have are:
\nObservations:
\nSo,
\nA. If we have images with label 10, the green-channel will have overlap with the red-channel, both of which are \"microtubule\"
\nB. If we have images with label 6, the green-channel will have overlap with the yellow-channel, both of which are \"endoplasmic reticulum\"
\nC. 1) If an image has neither label 10 nor 6, but do have red and yellow channel highlighting portions of the image, we can use the regions highlighted by the red and yellow for label 10 and 6 respectively.
Is this correct?
", "votes": 1, "replies": [ { "id": 1184158, "author_name": "UAxelsson", "author_url": "", "post_date": "2021-02-03T12:18:15.620000", "content": "A+B. You are correct, the red channel will help you to identify a microtubule pattern (label 10) in the green channel, and the yellow channel will help you identify an ER pattern (label 6) in the green channel. The reference channels (red/green/blue) can also help you to get an overview of the cell - certain patterns will always be distributed in a certain way in correlation to these. For example, labels 2 to 5 are all within the nucleus (i.e. the region of each cell with a signal in the blue channel).
\nC. All images have all the four channels, and signals from the markers (blue, yellow, red) are present in all cells in the image, independent of the green channel that you are classifying, in order to help you identify where the cells are, as well as where certain structures and regions within the cells are. This can in turn help you to segment the cells and to classify each cell to one or more label(s) according to the signal in the green channel. Note that each of the image-level labels can be present in all or in just a fraction of the individual cells in the image, and that some individual cells may also have additional labels, that are not mentioned among the image-level labels.
", "votes": 3, "replies": [] }, { "id": 1186834, "author_name": "Darek Kłeczek", "author_url": "", "post_date": "2021-02-05T04:35:37.817000", "content": "Hi @uaxelsson, can we ask a quick clarification on this part?:
\n\n\nsome individual cells may also have additional labels, that are not mentioned among the image-level labels.
\n
We understand that the additional label might be negative, but other than that it would mean that the image-level labels were not fully accurate and additional labels where added while tagging segmented cells, is that correct? Do we have an indication of how much the labels in the test set differ in this regard from the training set? @dschettler8845
", "votes": 2, "replies": [] }, { "id": 1187705, "author_name": "Darien Schettler", "author_url": "", "post_date": "2021-02-05T16:41:51.453000", "content": "I would like to see the answer to this as well. Is this only w.r.t. to the Negative label?
\nThanks for tagging me Darek!
", "votes": 0, "replies": [] }, { "id": 1187717, "author_name": "Emma Lundberg", "author_url": "", "post_date": "2021-02-05T16:51:32.180000", "content": "Yes there may be images where a single cell have a label that was neglected at the image level (for example a weak cytosolic pattern). Managing such weak-labels is part of this challenge, and it's impact is something we will analyze at the end of the challenge.
", "votes": 4, "replies": [] } ] }, { "id": 1179066, "author_name": "Konstantin Lopukhin", "author_url": "", "post_date": "2021-01-31T09:39:22.900000", "content": "From what I understand, some classes are more visible in other channels, e.g. Endoplasmic reticulum in yellow, see https://www.kaggle.com/lnhtrang/single-cell-patterns - other than that I think you are right that image-level classes can apply to a sub-set of the cells.
", "votes": 2, "replies": [ { "id": 1182088, "author_name": "UAxelsson", "author_url": "", "post_date": "2021-02-02T09:46:52.597000", "content": "Yes, but please note that it is always the pattern(s) in the green channel that you should classify. Yellow will show you what the Endoplasmic reticulum looks like in each cell. It is a marker present in all images, to help identify proteins of interest (in green) that also localize to the endoplasmic reticulum.
", "votes": 3, "replies": [] }, { "id": 1207467, "author_name": "Ayush Thakur", "author_url": "", "post_date": "2021-02-17T21:03:26.723000", "content": "Is it safe to assume that segmented cells without any signal from the green channel will have a negative(18) label with confidence 1? @UAxelsson
", "votes": 0, "replies": [] }, { "id": 1211560, "author_name": "Mohammed Amine Triki", "author_url": "", "post_date": "2021-02-20T10:29:14.037000", "content": "@uaxelsson does this mean that we should train models only on green images, when you say \"should classify\" ?
", "votes": 0, "replies": [] }, { "id": 1213899, "author_name": "Casper Winsnes", "author_url": "", "post_date": "2021-02-22T12:26:19.610000", "content": "@mohammedaminetriki The green channel is the channel for the protein of interest. What we want to know is what cellular compartments the protein is localizing to.
\nThe channels other than green are our reference points when annotating so they will probably help your model but you are in no way forced to use them.
Wow, that clarified a ton!
\nThanks for the detailed response @dschettler8845 and @lopuhin !!