\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F862111%2F4ce628e44b89d346115b9b7748b51ca5%2FScreenshot%20from%202023-02-10%2021-03-46.png?generation=1676063098860331&alt=media\")
\nignore marks etc, its just an example image
OK, I know I am hopeless if I am still talking about breast leterality, but still…
\nSome people talked about 9k+ laterality being wrong.
\nIn my thinking: as long as you are developing image level models and the cancer-labels of the images are correct, having 'wrong laterality' labels does not make a difference, does it?
\nAdditionally, if you try to correct the laterality, isn't there the possibility that you correct the laterality in test data, whereas the evaluation will be based on the original laterality of the images??
OK, I know I am hopeless if I am still talking about breast leterality, but still…
\nSome people talked about 9k+ laterality being wrong.
\nIn my thinking: as long as you are developing image level models and the cancer-labels of the images are correct, having 'wrong laterality' labels does not make a difference, does it?
\nAdditionally, if you try to correct the laterality, isn't there the possibility that you correct the laterality in test data, whereas the evaluation will be based on the original laterality of the images??
I believe the lateralities are correct because even the \"flipped\" lateralities have text on the image that is not flipped. I am not a radiologist but it could be different setups/preferences for each radiologist when taking the original mammogram. I think you are correct. So I don't think we can assume that laterality L means that the breast is on the left side of the image, it just means that the image is the left breast or right breast. I can't say whether the left or right breast could be used as a feature, I assume yes in the real world where breast cancer would be more common in the left breast but in this dataset, the cancer labels seem balanced.
", "rawMarkdown": "I believe the lateralities are correct because even the \"flipped\" lateralities have text on the image that is not flipped. I am not a radiologist but it could be different setups/preferences for each radiologist when taking the original mammogram. I think you are correct. So I don't think we can assume that laterality L means that the breast is on the left side of the image, it just means that the image is [the left breast or right breast](https://www.kaggle.com/competitions/rsna-breast-cancer-detection/data). I can't say whether the left or right breast could be used as a feature, I assume yes in the real world where breast cancer would be more common in the left breast but in this dataset, the cancer labels seem balanced.", "votes": 3 }, { "id": 2139682, "postDate": "2023-02-11T02:38:24.483Z", "content": "A script like this can be used to detect the actual laterality if both the csv and DICOM header give the wrong label:
\nsum_rows = np.(img, axis=)\nmidpoint = (np.floor((sum_rows) / ))\nmass_on_the_left_side = np.(sum_rows[:midpoint])\nmass_on_the_right_side = np.(sum_rows[midpoint + :])\n\n (laterality == mass_on_the_left_side < mass_on_the_right_side) (laterality == mass_on_the_left_side > mass_on_the_right_side):\n img = cv2.flip(img, flipCode=)\n\n\nIf you have augmentations and especially flip augmentation, then I am sure you are prepared for images having wrong laterality.
", "rawMarkdown": "A script like this can be used to detect the actual laterality if both the csv and DICOM header give the wrong label:\n\n```python\nsum_rows = np.sum(img, axis=0)\nmidpoint = int(np.floor(len(sum_rows) / 2.))\nmass_on_the_left_side = np.sum(sum_rows[0:midpoint])\nmass_on_the_right_side = np.sum(sum_rows[midpoint + 1:])\n\nif (laterality == 'L' and mass_on_the_left_side < mass_on_the_right_side) or (laterality == 'R' and mass_on_the_left_side > mass_on_the_right_side):\n img = cv2.flip(img, flipCode=1)\n\n# and so forth...\n```\nIf you have augmentations and especially flip augmentation, then I am sure you are prepared for images having wrong laterality.", "votes": 1 }, { "id": 2139546, "postDate": "2023-02-10T21:05:45.607Z", "content": "Afaik, professional examination prefers this order btw:
\n
\nignore marks etc, its just an example image
For me, laterality refers to the side of the human body where the breast is located, not the side on which the breast appears in a mammogram. The laterality does not depend on the position in which the mammogram was taken.
", "rawMarkdown": "For me, laterality refers to the side of the human body where the breast is located, not the side on which the breast appears in a mammogram. The laterality does not depend on the position in which the mammogram was taken.", "votes": 2 }, { "id": 2143446, "postDate": "2023-02-14T09:09:49.363Z", "content": "They are various opinions on the subjects. I don't think it is where the breast is located as the laterality can be both.
\nHere are some link.
\nhttps://patents.google.com/patent/US7796793B2/en
\nhttps://dicom.innolitics.com/ciods/digital-mammography-x-ray-image/mammography-image/00200062
\nhttps://dicom.nema.org/medical/dicom/2014c/output/chtml/part03/sect_C.8.11.7.html
If anyone got something clearer please share
", "rawMarkdown": "They are various opinions on the subjects. I don't think it is where the breast is located as the laterality can be both.\nHere are some link. \nhttps://patents.google.com/patent/US7796793B2/en\nhttps://dicom.innolitics.com/ciods/digital-mammography-x-ray-image/mammography-image/00200062\nhttps://dicom.nema.org/medical/dicom/2014c/output/chtml/part03/sect_C.8.11.7.html\n\nIf anyone got something clearer please share" } ], "comments": [ { "id": 2138212, "author_name": "outwrest", "author_url": "", "post_date": "2023-02-10T16:35:46.363000", "content": "I believe the lateralities are correct because even the \"flipped\" lateralities have text on the image that is not flipped. I am not a radiologist but it could be different setups/preferences for each radiologist when taking the original mammogram. I think you are correct. So I don't think we can assume that laterality L means that the breast is on the left side of the image, it just means that the image is the left breast or right breast. I can't say whether the left or right breast could be used as a feature, I assume yes in the real world where breast cancer would be more common in the left breast but in this dataset, the cancer labels seem balanced.
", "votes": 3, "replies": [] }, { "id": 2139682, "author_name": "Antti Isosalo", "author_url": "", "post_date": "2023-02-11T02:38:24.483000", "content": "A script like this can be used to detect the actual laterality if both the csv and DICOM header give the wrong label:
\nsum_rows = np.(img, axis=)\nmidpoint = (np.floor((sum_rows) / ))\nmass_on_the_left_side = np.(sum_rows[:midpoint])\nmass_on_the_right_side = np.(sum_rows[midpoint + :])\n\n (laterality == mass_on_the_left_side < mass_on_the_right_side) (laterality == mass_on_the_left_side > mass_on_the_right_side):\n img = cv2.flip(img, flipCode=)\n\n\nIf you have augmentations and especially flip augmentation, then I am sure you are prepared for images having wrong laterality.
", "votes": 1, "replies": [] }, { "id": 2139546, "author_name": "Eleftherios Fanioudakis", "author_url": "", "post_date": "2023-02-10T21:05:45.607000", "content": "Afaik, professional examination prefers this order btw:
\n
\nignore marks etc, its just an example image
For me, laterality refers to the side of the human body where the breast is located, not the side on which the breast appears in a mammogram. The laterality does not depend on the position in which the mammogram was taken.
", "votes": 2, "replies": [] }, { "id": 2143446, "author_name": "Koubouratou IDJATON", "author_url": "", "post_date": "2023-02-14T09:09:49.363000", "content": "They are various opinions on the subjects. I don't think it is where the breast is located as the laterality can be both.
\nHere are some link.
\nhttps://patents.google.com/patent/US7796793B2/en
\nhttps://dicom.innolitics.com/ciods/digital-mammography-x-ray-image/mammography-image/00200062
\nhttps://dicom.nema.org/medical/dicom/2014c/output/chtml/part03/sect_C.8.11.7.html
If anyone got something clearer please share
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2138199": "OK, I know I am hopeless if I am still talking about breast leterality, but still...\nSome people talked about 9k+ laterality being wrong. \nIn my thinking: as long as you are developing image level models and the cancer-labels of the images are correct, having 'wrong laterality' labels does not make a difference, does it? \nAdditionally, if you try to correct the laterality, isn't there the possibility that you correct the laterality in test data, whereas the evaluation will be based on the original laterality of the images??\n", "2138212": "I believe the lateralities are correct because even the \"flipped\" lateralities have text on the image that is not flipped. I am not a radiologist but it could be different setups/preferences for each radiologist when taking the original mammogram. I think you are correct. So I don't think we can assume that laterality L means that the breast is on the left side of the image, it just means that the image is [the left breast or right breast](https://www.kaggle.com/competitions/rsna-breast-cancer-detection/data). I can't say whether the left or right breast could be used as a feature, I assume yes in the real world where breast cancer would be more common in the left breast but in this dataset, the cancer labels seem balanced.", "2139682": "A script like this can be used to detect the actual laterality if both the csv and DICOM header give the wrong label:\n\n```python\nsum_rows = np.sum(img, axis=0)\nmidpoint = int(np.floor(len(sum_rows) / 2.))\nmass_on_the_left_side = np.sum(sum_rows[0:midpoint])\nmass_on_the_right_side = np.sum(sum_rows[midpoint + 1:])\n\nif (laterality == 'L' and mass_on_the_left_side < mass_on_the_right_side) or (laterality == 'R' and mass_on_the_left_side > mass_on_the_right_side):\n img = cv2.flip(img, flipCode=1)\n\n# and so forth...\n```\nIf you have augmentations and especially flip augmentation, then I am sure you are prepared for images having wrong laterality.", "2139546": "Afaik, professional examination prefers this order btw:\n\nignore marks etc, its just an example image", "2139543": "For me, laterality refers to the side of the human body where the breast is located, not the side on which the breast appears in a mammogram. The laterality does not depend on the position in which the mammogram was taken.", "2143446": "They are various opinions on the subjects. I don't think it is where the breast is located as the laterality can be both.\nHere are some link. \nhttps://patents.google.com/patent/US7796793B2/en\nhttps://dicom.innolitics.com/ciods/digital-mammography-x-ray-image/mammography-image/00200062\nhttps://dicom.nema.org/medical/dicom/2014c/output/chtml/part03/sect_C.8.11.7.html\n\nIf anyone got something clearer please share" } }