![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F898402%2F3205f8fe28398cb65930748e01c291ec%2FBiradScore.png?generation=1669774326334573&alt=media\")
One of the training columns is called BIRADS (Breast Imaging Reporting and Database System).
\nBIRADS is a scale that radiologists use to roughly specify the level of abnormality as related to breast cancer. The range is 0-6 with zero being normal, and 6 being confirmed cancer (by biopsy).
\nPatients with a 6 BIRAD definitely have cancer. 5's probably have cancer and 1-4 have some amount of abnormality.
\nHere's a good explanation of it -> https://www.healthline.com/health/birads-score#scoring
", "messages": [ { "id": 2047908, "postDate": "2022-11-29T04:59:34.907Z", "content": "One of the training columns is called BIRADS (Breast Imaging Reporting and Database System).
\nBIRADS is a scale that radiologists use to roughly specify the level of abnormality as related to breast cancer. The range is 0-6 with zero being normal, and 6 being confirmed cancer (by biopsy).
\nPatients with a 6 BIRAD definitely have cancer. 5's probably have cancer and 1-4 have some amount of abnormality.
\nHere's a good explanation of it -> https://www.healthline.com/health/birads-score#scoring
", "rawMarkdown": "One of the training columns is called BIRADS (Breast Imaging Reporting and Database System).\n\nBIRADS is a scale that radiologists use to roughly specify the level of abnormality as related to breast cancer. The range is 0-6 with zero being normal, and 6 being confirmed cancer (by biopsy). \n\nPatients with a 6 BIRAD definitely have cancer. 5's probably have cancer and 1-4 have some amount of abnormality.\n\nHere's a good explanation of it -> https://www.healthline.com/health/birads-score#scoring", "votes": 23 }, { "id": 2052961, "postDate": "2022-12-02T17:16:30.997Z", "content": "To try and make things clearer:
\nThe data provided is from Screening exams. That means the patient has no known issues, and is coming in for a routine (annual or bi-annual) exam. Typical recommendations are to start at 40 years old or 50 years old, with variations by country and medical society. Patients with a significant family history or chest radiation exposure might start earlier.
\nThe Bi-Rads is assigned to a Screening Exam BEFORE the real answer is known through a biopsy. It is the initial evaluation of the Images, not the final result.
\nThe Screening exam is classified in 3 possible ways. I have not looked closely to see if the data provided gives a Bi-Rads for each Patient or each Breast.
\n0 - more imaging is needed. That means something is suspicious for cancer and needs either more Mammographic images, an US and/or an MRI.
\n1 - Negative. Nothing to comment on. Continue routine screening.
\n2 - Benign - something to comment on, but nothing that is suspicious for cancer. For instance, vascular calcifications that might relate to cardiovascular disease, but have nothing to do with cancer. Benign (non-cancerous) appearing lymph nodes. There could be a nodule that has been there for multiple years and so is not suspicious. Unfortunately, we don't have multiple years of study, so we cannot tell if something has been there for years. I do not know if the readers who assigned the Bi-Rads had the prior exams available (the answer might be given in the Overview). Continue routine screening.
You are not supposed to use the other Bi-Rads categories for a Screening Exam. Even if something is obviously cancer, you are still supposed to do additional imaging before sending the patient for biopsy. I didn't check if the provided data follows this rule.
\nAfter you have additional imaging (More mammograms images (compressed images, magnified images, different posiitionin), Ultrasound or MRI):
\n2 - Benign. With the help of more imaging, we've explained what was concerning and are no longer concerned. Go back to routine Screening.
\n3 - Probably Benign. Typically less than a 2% chance of there being cancer. Usually followup with a Mammogram or other study in 6 months.
\n4 - Suspicious Malignant - something is seen that needs a biopsy, but might not be cancer. Biopsy recommended. In practice, some of these patients end up being followed in 6 months (like a Bi-Rads 3) rather than getting a biopsy.
\n5 - Malignant - something is seen that is almost certainly cancer. Needs a biopsy.
\n6 - Proven Malignant. Used after biopsy in evaluating the breast for other areas of concern, checking the results of surgery, or other reasons.
A few additional notes:
\nI did not check if the data only includes \"born Female\" patients. There are Screening guidelines for non Binary patients (an article was posted in another thread).
After a patient has a known, treated cancer, they usually have a \"Diagnostic\" exam for a few years instead of a routine Screening exam. But after a few years, they might go back to routine annual screening. I don't know if these (previous cancer) patients are included in the data.
\nSome patients will have surgical clips from previous surgery - which might or might not have been cancer. I have not looked at all the images to see if there are clips from prior surgery.
\nSince a cancer could have been missed by the radiologist, usually data for this type of contest would have followed the patient for at least 2 years to make sure a cancer didn't show up. I don't know if they did that for this data.
\nCurrent technology now includes Tomosynthesis. Basically, instead of just a single image of each breast in each projection, we take about 20-40 images in each projection. Sort of like a CT scan. In the United States at least, this is becoming the standard of care. It lets you see much more detail in the breast and you can tell more accurately where in the breast something is. The datasets are much larger and make training a model much harder.
\nSince we take two images of each breast, if you see the same abnormality in both images, you can be more confident that it is a real finding than if you see it in only one image. So the best models would use both images.
\nBreast density (scale of 1 to 4) has two impacts on determining if there is cancer. First, it is harder to see cancer in denser breasts. Separately, having denser breasts is itself a risk factor for cancer - not just that it is harder to see, but that it is more likely to be there. Usually the breast density is determined by the radiologist just looking at the film, but there are AI programs that automate the density determination.
\nI hope this helps.
\n-Rich
", "rawMarkdown": "To try and make things clearer:\n\nThe data provided is from Screening exams. That means the patient has no known issues, and is coming in for a routine (annual or bi-annual) exam. Typical recommendations are to start at 40 years old or 50 years old, with variations by country and medical society. Patients with a significant family history or chest radiation exposure might start earlier.\n\nThe Bi-Rads is assigned to a Screening Exam BEFORE the real answer is known through a biopsy. It is the initial evaluation of the Images, not the final result.\n\nThe Screening exam is classified in 3 possible ways. I have not looked closely to see if the data provided gives a Bi-Rads for each Patient or each Breast.\n\n0 - more imaging is needed. That means something is suspicious for cancer and needs either more Mammographic images, an US and/or an MRI.\n1 - Negative. Nothing to comment on. Continue routine screening.\n2 - Benign - something to comment on, but nothing that is suspicious for cancer. For instance, vascular calcifications that might relate to cardiovascular disease, but have nothing to do with cancer. Benign (non-cancerous) appearing lymph nodes. There could be a nodule that has been there for multiple years and so is not suspicious. Unfortunately, we don't have multiple years of study, so we cannot tell if something has been there for years. I do not know if the readers who assigned the Bi-Rads had the prior exams available (the answer might be given in the Overview). Continue routine screening.\n\nYou are not supposed to use the other Bi-Rads categories for a Screening Exam. Even if something is obviously cancer, you are still supposed to do additional imaging before sending the patient for biopsy. I didn't check if the provided data follows this rule.\n\nAfter you have additional imaging (More mammograms images (compressed images, magnified images, different posiitionin), Ultrasound or MRI):\n\n2 - Benign. With the help of more imaging, we've explained what was concerning and are no longer concerned. Go back to routine Screening.\n3 - Probably Benign. Typically less than a 2% chance of there being cancer. Usually followup with a Mammogram or other study in 6 months.\n4 - Suspicious Malignant - something is seen that needs a biopsy, but might not be cancer. Biopsy recommended. In practice, some of these patients end up being followed in 6 months (like a Bi-Rads 3) rather than getting a biopsy.\n5 - Malignant - something is seen that is almost certainly cancer. Needs a biopsy.\n6 - Proven Malignant. Used after biopsy in evaluating the breast for other areas of concern, checking the results of surgery, or other reasons.\n\nA few additional notes:\nI did not check if the data only includes \"born Female\" patients. There are Screening guidelines for non Binary patients (an article was posted in another thread).\n\nAfter a patient has a known, treated cancer, they usually have a \"Diagnostic\" exam for a few years instead of a routine Screening exam. But after a few years, they might go back to routine annual screening. I don't know if these (previous cancer) patients are included in the data.\n\nSome patients will have surgical clips from previous surgery - which might or might not have been cancer. I have not looked at all the images to see if there are clips from prior surgery.\n\nSince a cancer could have been missed by the radiologist, usually data for this type of contest would have followed the patient for at least 2 years to make sure a cancer didn't show up. I don't know if they did that for this data.\n\nCurrent technology now includes Tomosynthesis. Basically, instead of just a single image of each breast in each projection, we take about 20-40 images in each projection. Sort of like a CT scan. In the United States at least, this is becoming the standard of care. It lets you see much more detail in the breast and you can tell more accurately where in the breast something is. The datasets are much larger and make training a model much harder.\n\nSince we take two images of each breast, if you see the same abnormality in both images, you can be more confident that it is a real finding than if you see it in only one image. So the best models would use both images.\n\nBreast density (scale of 1 to 4) has two impacts on determining if there is cancer. First, it is harder to see cancer in denser breasts. Separately, having denser breasts is itself a risk factor for cancer - not just that it is harder to see, but that it is more likely to be there. Usually the breast density is determined by the radiologist just looking at the film, but there are AI programs that automate the density determination.\n\nI hope this helps.\n\n-Rich", "votes": 4 }, { "id": 2049226, "postDate": "2022-11-30T02:12:34.607Z", "content": "BIRAD Classification System for quick reference
\n
I was confused about this column.
\nThank you @davidbroberts for this share and explanation.
@davidbroberts It is confusing in the train dataset.
\nAll the images with a cancer label 1 have a BIRAD score of 0. What does this mean?
Having the \"cancer\" target makes BIRADS 5 and 6 kind of redundant. It wouldn't make sense to have a cancer=1 and BIRADS=5 in the same row since cancer = 1 should be the same as BIRADS = 6. I suspect the hosts didn't include BIRADS on the cancerous images to avoid confusion.
\nI think BIRADS will only be useful to group the non cancerous images.
", "rawMarkdown": "Having the \"cancer\" target makes BIRADS 5 and 6 kind of redundant. It wouldn't make sense to have a cancer=1 and BIRADS=5 in the same row since cancer = 1 should be the same as BIRADS = 6. I suspect the hosts didn't include BIRADS on the cancerous images to avoid confusion.\n\nI think BIRADS will only be useful to group the non cancerous images." }, { "id": 2049376, "postDate": "2022-11-30T05:06:52.597Z", "content": "The images used in this dataset are all from screening exams only. For this competition BIRADS 0 indicates that the radiologist reviewing the mammograms had the patient come back for other follow up procedures such as as diagnostic mammograms and biopsies to confirm the presence/absence of cancer and update their BIRADS score.
", "rawMarkdown": "The images used in this dataset are all from screening exams only. For this competition BIRADS 0 indicates that the radiologist reviewing the mammograms had the patient come back for other follow up procedures such as as diagnostic mammograms and biopsies to confirm the presence/absence of cancer and update their BIRADS score.", "votes": 3 } ] }, { "id": 2048241, "postDate": "2022-11-29T09:59:53.020Z", "content": "Can you clarify one thing, as stated in \"Data\". What does it mean by \"2\"? Isn't \"1\" negative is equal to \"2\" (normal)?
\n\n\nBIRADS - 0 if the breast required follow-up, 1 if the breast was rated as negative for cancer, and 2 if the breast was rated as normal. Only provided for train.
\n
Isn't \"negative\" equals \"normal\"?
", "rawMarkdown": "Can you clarify one thing, as stated in \"Data\". What does it mean by \"2\"? Isn't \"1\" negative is equal to \"2\" (normal)? \n\n> BIRADS - 0 if the breast required follow-up, 1 if the breast was rated as negative for cancer, and 2 if the breast was rated as normal. Only provided for train.\n\nIsn't \"negative\" equals \"normal\"?", "replies": [ { "id": 2048531, "postDate": "2022-11-29T14:16:26.353Z", "content": "In this case, Negative and Normal aren't the same. Negative means \"no cancer\" but not necessarily a \"normal\" appearing breast. The breast could have other abnormalities that require follow up, but do not require increasing the BIRADS score.
", "rawMarkdown": "In this case, Negative and Normal aren't the same. Negative means \"no cancer\" but not necessarily a \"normal\" appearing breast. The breast could have other abnormalities that require follow up, but do not require increasing the BIRADS score.", "votes": 1 }, { "id": 2048544, "postDate": "2022-11-29T14:22:30.497Z", "content": "But, in regards to this competition isn't it irrelevant? We just need to consider the 2 classification \"cancer\" or \"not\", no?
", "rawMarkdown": "But, in regards to this competition isn't it irrelevant? We just need to consider the 2 classification \"cancer\" or \"not\", no?" }, { "id": 2048552, "postDate": "2022-11-29T14:30:18.343Z", "content": "BIRADS category 2 means \"abnormal\" but not cancerous. The only BIRADS score that actually means \"positive for cancer\" is 6. That's because cancer isn't diagnosed from images, it's only after a biopsy that cancer can be diagnosed.
\nBIRADS 5 means the doctor is pretty certain the abnormality is cancerous based on it's appearance (and other clinical data), but, without a biopsy, it isn't proven. If you get a BIRADS 5 score on your mammogram, it's pretty much certain that you have a cancerous tumor.
", "rawMarkdown": "BIRADS category 2 means \"abnormal\" but not cancerous. The only BIRADS score that actually means \"positive for cancer\" is 6. That's because cancer isn't diagnosed from images, it's only after a biopsy that cancer can be diagnosed.\n\nBIRADS 5 means the doctor is pretty certain the abnormality is cancerous based on it's appearance (and other clinical data), but, without a biopsy, it isn't proven. If you get a BIRADS 5 score on your mammogram, it's pretty much certain that you have a cancerous tumor.", "votes": 3 }, { "id": 2048565, "postDate": "2022-11-29T14:42:15.170Z", "content": "Ohhh, okay. I understand, thank you. I got it now, thank you.
", "rawMarkdown": "Ohhh, okay. I understand, thank you. I got it now, thank you." } ] }, { "id": 2047958, "postDate": "2022-11-29T06:19:41.880Z", "rawMarkdown": "", "votes": -1, "isDeleted": true } ], "comments": [ { "id": 2052961, "author_name": "quadcore/Richard Epstein", "author_url": "", "post_date": "2022-12-02T17:16:30.997000", "content": "To try and make things clearer:
\nThe data provided is from Screening exams. That means the patient has no known issues, and is coming in for a routine (annual or bi-annual) exam. Typical recommendations are to start at 40 years old or 50 years old, with variations by country and medical society. Patients with a significant family history or chest radiation exposure might start earlier.
\nThe Bi-Rads is assigned to a Screening Exam BEFORE the real answer is known through a biopsy. It is the initial evaluation of the Images, not the final result.
\nThe Screening exam is classified in 3 possible ways. I have not looked closely to see if the data provided gives a Bi-Rads for each Patient or each Breast.
\n0 - more imaging is needed. That means something is suspicious for cancer and needs either more Mammographic images, an US and/or an MRI.
\n1 - Negative. Nothing to comment on. Continue routine screening.
\n2 - Benign - something to comment on, but nothing that is suspicious for cancer. For instance, vascular calcifications that might relate to cardiovascular disease, but have nothing to do with cancer. Benign (non-cancerous) appearing lymph nodes. There could be a nodule that has been there for multiple years and so is not suspicious. Unfortunately, we don't have multiple years of study, so we cannot tell if something has been there for years. I do not know if the readers who assigned the Bi-Rads had the prior exams available (the answer might be given in the Overview). Continue routine screening.
You are not supposed to use the other Bi-Rads categories for a Screening Exam. Even if something is obviously cancer, you are still supposed to do additional imaging before sending the patient for biopsy. I didn't check if the provided data follows this rule.
\nAfter you have additional imaging (More mammograms images (compressed images, magnified images, different posiitionin), Ultrasound or MRI):
\n2 - Benign. With the help of more imaging, we've explained what was concerning and are no longer concerned. Go back to routine Screening.
\n3 - Probably Benign. Typically less than a 2% chance of there being cancer. Usually followup with a Mammogram or other study in 6 months.
\n4 - Suspicious Malignant - something is seen that needs a biopsy, but might not be cancer. Biopsy recommended. In practice, some of these patients end up being followed in 6 months (like a Bi-Rads 3) rather than getting a biopsy.
\n5 - Malignant - something is seen that is almost certainly cancer. Needs a biopsy.
\n6 - Proven Malignant. Used after biopsy in evaluating the breast for other areas of concern, checking the results of surgery, or other reasons.
A few additional notes:
\nI did not check if the data only includes \"born Female\" patients. There are Screening guidelines for non Binary patients (an article was posted in another thread).
After a patient has a known, treated cancer, they usually have a \"Diagnostic\" exam for a few years instead of a routine Screening exam. But after a few years, they might go back to routine annual screening. I don't know if these (previous cancer) patients are included in the data.
\nSome patients will have surgical clips from previous surgery - which might or might not have been cancer. I have not looked at all the images to see if there are clips from prior surgery.
\nSince a cancer could have been missed by the radiologist, usually data for this type of contest would have followed the patient for at least 2 years to make sure a cancer didn't show up. I don't know if they did that for this data.
\nCurrent technology now includes Tomosynthesis. Basically, instead of just a single image of each breast in each projection, we take about 20-40 images in each projection. Sort of like a CT scan. In the United States at least, this is becoming the standard of care. It lets you see much more detail in the breast and you can tell more accurately where in the breast something is. The datasets are much larger and make training a model much harder.
\nSince we take two images of each breast, if you see the same abnormality in both images, you can be more confident that it is a real finding than if you see it in only one image. So the best models would use both images.
\nBreast density (scale of 1 to 4) has two impacts on determining if there is cancer. First, it is harder to see cancer in denser breasts. Separately, having denser breasts is itself a risk factor for cancer - not just that it is harder to see, but that it is more likely to be there. Usually the breast density is determined by the radiologist just looking at the film, but there are AI programs that automate the density determination.
\nI hope this helps.
\n-Rich
", "votes": 4, "replies": [] }, { "id": 2049226, "author_name": "gradientBoost", "author_url": "", "post_date": "2022-11-30T02:12:34.607000", "content": "BIRAD Classification System for quick reference
\n
I was confused about this column.
\nThank you @davidbroberts for this share and explanation.
@davidbroberts It is confusing in the train dataset.
\nAll the images with a cancer label 1 have a BIRAD score of 0. What does this mean?
Having the \"cancer\" target makes BIRADS 5 and 6 kind of redundant. It wouldn't make sense to have a cancer=1 and BIRADS=5 in the same row since cancer = 1 should be the same as BIRADS = 6. I suspect the hosts didn't include BIRADS on the cancerous images to avoid confusion.
\nI think BIRADS will only be useful to group the non cancerous images.
", "votes": 0, "replies": [] }, { "id": 2049376, "author_name": "Sohier Dane", "author_url": "", "post_date": "2022-11-30T05:06:52.597000", "content": "The images used in this dataset are all from screening exams only. For this competition BIRADS 0 indicates that the radiologist reviewing the mammograms had the patient come back for other follow up procedures such as as diagnostic mammograms and biopsies to confirm the presence/absence of cancer and update their BIRADS score.
", "votes": 3, "replies": [] } ] }, { "id": 2048241, "author_name": "Hey24sheep", "author_url": "", "post_date": "2022-11-29T09:59:53.020000", "content": "Can you clarify one thing, as stated in \"Data\". What does it mean by \"2\"? Isn't \"1\" negative is equal to \"2\" (normal)?
\n\n\nBIRADS - 0 if the breast required follow-up, 1 if the breast was rated as negative for cancer, and 2 if the breast was rated as normal. Only provided for train.
\n
Isn't \"negative\" equals \"normal\"?
", "votes": 0, "replies": [ { "id": 2048531, "author_name": "David Roberts", "author_url": "", "post_date": "2022-11-29T14:16:26.353000", "content": "In this case, Negative and Normal aren't the same. Negative means \"no cancer\" but not necessarily a \"normal\" appearing breast. The breast could have other abnormalities that require follow up, but do not require increasing the BIRADS score.
", "votes": 1, "replies": [] }, { "id": 2048544, "author_name": "Hey24sheep", "author_url": "", "post_date": "2022-11-29T14:22:30.497000", "content": "But, in regards to this competition isn't it irrelevant? We just need to consider the 2 classification \"cancer\" or \"not\", no?
", "votes": 0, "replies": [] }, { "id": 2048552, "author_name": "David Roberts", "author_url": "", "post_date": "2022-11-29T14:30:18.343000", "content": "BIRADS category 2 means \"abnormal\" but not cancerous. The only BIRADS score that actually means \"positive for cancer\" is 6. That's because cancer isn't diagnosed from images, it's only after a biopsy that cancer can be diagnosed.
\nBIRADS 5 means the doctor is pretty certain the abnormality is cancerous based on it's appearance (and other clinical data), but, without a biopsy, it isn't proven. If you get a BIRADS 5 score on your mammogram, it's pretty much certain that you have a cancerous tumor.
", "votes": 3, "replies": [] }, { "id": 2048565, "author_name": "Hey24sheep", "author_url": "", "post_date": "2022-11-29T14:42:15.170000", "content": "Ohhh, okay. I understand, thank you. I got it now, thank you.
", "votes": 0, "replies": [] } ] }, { "id": 2047958, "author_name": "", "author_url": "", "post_date": "2022-11-29T06:19:41.880000", "content": "", "votes": -1, "replies": [] } ], "raw_markdown_by_id": { "2047908": "One of the training columns is called BIRADS (Breast Imaging Reporting and Database System).\n\nBIRADS is a scale that radiologists use to roughly specify the level of abnormality as related to breast cancer. The range is 0-6 with zero being normal, and 6 being confirmed cancer (by biopsy). \n\nPatients with a 6 BIRAD definitely have cancer. 5's probably have cancer and 1-4 have some amount of abnormality.\n\nHere's a good explanation of it -> https://www.healthline.com/health/birads-score#scoring", "2052961": "To try and make things clearer:\n\nThe data provided is from Screening exams. That means the patient has no known issues, and is coming in for a routine (annual or bi-annual) exam. Typical recommendations are to start at 40 years old or 50 years old, with variations by country and medical society. Patients with a significant family history or chest radiation exposure might start earlier.\n\nThe Bi-Rads is assigned to a Screening Exam BEFORE the real answer is known through a biopsy. It is the initial evaluation of the Images, not the final result.\n\nThe Screening exam is classified in 3 possible ways. I have not looked closely to see if the data provided gives a Bi-Rads for each Patient or each Breast.\n\n0 - more imaging is needed. That means something is suspicious for cancer and needs either more Mammographic images, an US and/or an MRI.\n1 - Negative. Nothing to comment on. Continue routine screening.\n2 - Benign - something to comment on, but nothing that is suspicious for cancer. For instance, vascular calcifications that might relate to cardiovascular disease, but have nothing to do with cancer. Benign (non-cancerous) appearing lymph nodes. There could be a nodule that has been there for multiple years and so is not suspicious. Unfortunately, we don't have multiple years of study, so we cannot tell if something has been there for years. I do not know if the readers who assigned the Bi-Rads had the prior exams available (the answer might be given in the Overview). Continue routine screening.\n\nYou are not supposed to use the other Bi-Rads categories for a Screening Exam. Even if something is obviously cancer, you are still supposed to do additional imaging before sending the patient for biopsy. I didn't check if the provided data follows this rule.\n\nAfter you have additional imaging (More mammograms images (compressed images, magnified images, different posiitionin), Ultrasound or MRI):\n\n2 - Benign. With the help of more imaging, we've explained what was concerning and are no longer concerned. Go back to routine Screening.\n3 - Probably Benign. Typically less than a 2% chance of there being cancer. Usually followup with a Mammogram or other study in 6 months.\n4 - Suspicious Malignant - something is seen that needs a biopsy, but might not be cancer. Biopsy recommended. In practice, some of these patients end up being followed in 6 months (like a Bi-Rads 3) rather than getting a biopsy.\n5 - Malignant - something is seen that is almost certainly cancer. Needs a biopsy.\n6 - Proven Malignant. Used after biopsy in evaluating the breast for other areas of concern, checking the results of surgery, or other reasons.\n\nA few additional notes:\nI did not check if the data only includes \"born Female\" patients. There are Screening guidelines for non Binary patients (an article was posted in another thread).\n\nAfter a patient has a known, treated cancer, they usually have a \"Diagnostic\" exam for a few years instead of a routine Screening exam. But after a few years, they might go back to routine annual screening. I don't know if these (previous cancer) patients are included in the data.\n\nSome patients will have surgical clips from previous surgery - which might or might not have been cancer. I have not looked at all the images to see if there are clips from prior surgery.\n\nSince a cancer could have been missed by the radiologist, usually data for this type of contest would have followed the patient for at least 2 years to make sure a cancer didn't show up. I don't know if they did that for this data.\n\nCurrent technology now includes Tomosynthesis. Basically, instead of just a single image of each breast in each projection, we take about 20-40 images in each projection. Sort of like a CT scan. In the United States at least, this is becoming the standard of care. It lets you see much more detail in the breast and you can tell more accurately where in the breast something is. The datasets are much larger and make training a model much harder.\n\nSince we take two images of each breast, if you see the same abnormality in both images, you can be more confident that it is a real finding than if you see it in only one image. So the best models would use both images.\n\nBreast density (scale of 1 to 4) has two impacts on determining if there is cancer. First, it is harder to see cancer in denser breasts. Separately, having denser breasts is itself a risk factor for cancer - not just that it is harder to see, but that it is more likely to be there. Usually the breast density is determined by the radiologist just looking at the film, but there are AI programs that automate the density determination.\n\nI hope this helps.\n\n-Rich", "2049226": "BIRAD Classification System for quick reference\n\n", "2049246": "I was confused about this column.\nThank you @davidbroberts for this share and explanation.", "2049222": "@davidbroberts It is confusing in the train dataset. \nAll the images with a cancer label 1 have a BIRAD score of 0. What does this mean?", "2048241": "Can you clarify one thing, as stated in \"Data\". What does it mean by \"2\"? Isn't \"1\" negative is equal to \"2\" (normal)? \n\n> BIRADS - 0 if the breast required follow-up, 1 if the breast was rated as negative for cancer, and 2 if the breast was rated as normal. Only provided for train.\n\nIsn't \"negative\" equals \"normal\"?", "2047958": "" } }