![\"https://i.ibb.co/K5XY3gK/Selection-311.png\"](\"https://i.ibb.co/K5XY3gK/Selection-311.png\")
The Digital Database for Screening Mammography（DDSM）
\nHeath M, Bowyer K, Kopans D, et al. The Digital Database for Screening Mammography[C]// Proceedings of the Fifth International Workshop on Digital Mammography.Medical Physics Publishing. 2001: 212–218.
\nhttp://www.eng.usf.edu/cvprg/Mammography/Database.html
\nhttps://www.kaggle.com/datasets/cheddad/miniddsm2
The mammographic image analysis society digital mammogram database（MIAS）
\nSUCKLING J P. The mammographic image analysis society digital mammogram database[J]. Digital Mammo, 1994: 375–386.
\nhttps://www.repository.cam.ac.uk/handle/1810/250394
\nhttp://peipa.essex.ac.uk/info/mias.html
\nhttps://www.kaggle.com/datasets/kmader/mias-mammography
INbreast
\nMoreira I C, Amaral I, Domingues I, et al. Inbreast: toward a fullfield digital mammographic database[J]. Academic radiology, 2012, 19(2): 236–248.
\nhttps://pubmed.ncbi.nlm.nih.gov/22078258/
\nhttps://www.kaggle.com/datasets/martholi/inbreast
————————2023-1-2
\nhttps://www.kaggle.com/datasets/tommyngx/cmmd2022
The Digital Database for Screening Mammography（DDSM）
\nHeath M, Bowyer K, Kopans D, et al. The Digital Database for Screening Mammography[C]// Proceedings of the Fifth International Workshop on Digital Mammography.Medical Physics Publishing. 2001: 212–218.
\nhttp://www.eng.usf.edu/cvprg/Mammography/Database.html
\nhttps://www.kaggle.com/datasets/cheddad/miniddsm2
The mammographic image analysis society digital mammogram database（MIAS）
\nSUCKLING J P. The mammographic image analysis society digital mammogram database[J]. Digital Mammo, 1994: 375–386.
\nhttps://www.repository.cam.ac.uk/handle/1810/250394
\nhttp://peipa.essex.ac.uk/info/mias.html
\nhttps://www.kaggle.com/datasets/kmader/mias-mammography
INbreast
\nMoreira I C, Amaral I, Domingues I, et al. Inbreast: toward a fullfield digital mammographic database[J]. Academic radiology, 2012, 19(2): 236–248.
\nhttps://pubmed.ncbi.nlm.nih.gov/22078258/
\nhttps://www.kaggle.com/datasets/martholi/inbreast
————————2023-1-2
\nhttps://www.kaggle.com/datasets/tommyngx/cmmd2022
Great info! 🙂 As we are using models trained on Imagenet which contains images vastly different from the medical images in this competition, doing pretraining on other datasets (particularly ones that have rich labels) can go a long way! Thx for sharing!
", "rawMarkdown": "Great info! 🙂 As we are using models trained on Imagenet which contains images vastly different from the medical images in this competition, doing pretraining on other datasets (particularly ones that have rich labels) can go a long way! Thx for sharing!", "votes": 3, "replies": [ { "id": 2076905, "postDate": "2022-12-27T01:40:50.947Z", "content": "\"doing pretraining on other datasets (particularly ones that have rich labels) can go a long way!\"
\n
some dataset does closely resemble that of kaggle. in fact i suspect part of kaggle datset is part of NYU set or using the same scanner machine. check the paper details (e.g. biopsy label, BiRADS for 1,2,3 only) too!
\nwhile NYU set is not public, their pretrain models (several of them) are.
\nyou can use them
here is a paper that compares results with and without NYU pretrain model:
\n[1] Evaluation of deep learning-based artificial intelligence techniques for breast cancer detection on mammograms: Results from a retrospective study using a BreastScreen Victoria dataset
", "rawMarkdown": "\"doing pretraining on other datasets (particularly ones that have rich labels) can go a long way!\"\n\n\n\nsome dataset does closely resemble that of kaggle. in fact i suspect part of kaggle datset is part of NYU set or using the same scanner machine. check the paper details (e.g. biopsy label, BiRADS for 1,2,3 only) too!\n\nwhile NYU set is not public, their pretrain models (several of them) are.\nyou can use them\n\nhere is a paper that compares results with and without NYU pretrain model:\n\n[1] Evaluation of deep learning-based artificial intelligence techniques for breast cancer detection on mammograms: Results from a retrospective study using a BreastScreen Victoria dataset ", "votes": 7, "replies": [ { "id": 2077901, "postDate": "2022-12-28T00:04:23.647Z", "content": "very useful info @hengck23, thanks ofr sharing! 🙂
\nThe only problem here is that if we would use models that were pretrained on parts of this dataset, we might not be able to fully trust our validation results (as the model might have seen the image we have in our validation set already in train)
\nStill, that is a very useful piece of info! Thank you!
", "rawMarkdown": "very useful info @hengck23, thanks ofr sharing! 🙂 \n\nThe only problem here is that if we would use models that were pretrained on parts of this dataset, we might not be able to fully trust our validation results (as the model might have seen the image we have in our validation set already in train)\n\nStill, that is a very useful piece of info! Thank you!" }, { "id": 2077922, "postDate": "2022-12-28T00:24:51.650Z", "content": "@maggiemd @sohier
\nIs it true that NYU dataset is a part of the competition dataset (especially test dataset)?
\nIf so, using NYU pretrained model could cause a serious leakage, which is similar to the situation of cassava competition
Hello CHENJIE,
\nDo you know it this datatsets have any licence issues related to their use? by that I mean, could we infringe the database copyright if we win the competition and collect the reward by training our model with this datatset? or are they open for any kind of use?
Great info! 🙂 As we are using models trained on Imagenet which contains images vastly different from the medical images in this competition, doing pretraining on other datasets (particularly ones that have rich labels) can go a long way! Thx for sharing!
", "votes": 3, "replies": [ { "id": 2076905, "author_name": "hengck23", "author_url": "", "post_date": "2022-12-27T01:40:50.947000", "content": "\"doing pretraining on other datasets (particularly ones that have rich labels) can go a long way!\"
\n
some dataset does closely resemble that of kaggle. in fact i suspect part of kaggle datset is part of NYU set or using the same scanner machine. check the paper details (e.g. biopsy label, BiRADS for 1,2,3 only) too!
\nwhile NYU set is not public, their pretrain models (several of them) are.
\nyou can use them
here is a paper that compares results with and without NYU pretrain model:
\n[1] Evaluation of deep learning-based artificial intelligence techniques for breast cancer detection on mammograms: Results from a retrospective study using a BreastScreen Victoria dataset
", "votes": 7, "replies": [ { "id": 2077901, "author_name": "Radek Osmulski", "author_url": "", "post_date": "2022-12-28T00:04:23.647000", "content": "very useful info @hengck23, thanks ofr sharing! 🙂
\nThe only problem here is that if we would use models that were pretrained on parts of this dataset, we might not be able to fully trust our validation results (as the model might have seen the image we have in our validation set already in train)
\nStill, that is a very useful piece of info! Thank you!
", "votes": 0, "replies": [] }, { "id": 2077922, "author_name": "tomoo inubushi", "author_url": "", "post_date": "2022-12-28T00:24:51.650000", "content": "@maggiemd @sohier
\nIs it true that NYU dataset is a part of the competition dataset (especially test dataset)?
\nIf so, using NYU pretrained model could cause a serious leakage, which is similar to the situation of cassava competition
Hello CHENJIE,
\nDo you know it this datatsets have any licence issues related to their use? by that I mean, could we infringe the database copyright if we win the competition and collect the reward by training our model with this datatset? or are they open for any kind of use?