I saw them on Mayo Clinic - STRIP AI competition too. Why do they occur? Are they low quality versions of the tiff pyramidal images? Do they also exist on raw data or is it some kind of an artifact that happens while exporting the data to Kaggle?
", "messages": [ { "id": 2489524, "postDate": "2023-10-20T04:01:01.423Z", "content": "During histopathology tissue processing, very thin slices are cut from the wax tissue block. Since glass slides are rectangular, they can usually fit 2 slices from the block. If the tissue in the wax block in small and thin, they can fit even more slices.
\nThen probably depending on the slide scanner, scanning settings, and annotator, the saved image is either the whole glass slide, or the best representative slice.
\nIn the training dataset, I think the most in one slide is 6 slices.
\nI vaguely remember reading an article or study about whether to use all the slices vs just one slice for model training. I think they concluded that it's more cost effective or efficient to use one of the slices, since there are only subtle differences.
I thought they were artificial copies of each other. How can they cut slices that look almost identical?
", "rawMarkdown": "I thought they were artificial copies of each other. How can they cut slices that look almost identical?", "replies": [ { "id": 2489916, "postDate": "2023-10-20T09:42:19.940Z", "content": "Here is a link to how tissue processing works https://www.leicabiosystems.com/knowledge-pathway/an-introduction-to-specimen-processing/.
\nThey look the same since the cut slices are less than 2 micrometers thick.
", "rawMarkdown": "Here is a link to how tissue processing works [https://www.leicabiosystems.com/knowledge-pathway/an-introduction-to-specimen-processing/](url).\n\nThey look the same since the cut slices are less than 2 micrometers thick.", "votes": 7 }, { "id": 2527818, "postDate": "2023-11-16T20:30:58.050Z", "content": "Hi @nolialonso All the slice present in single images are of same category ? If this is the scenario we can randomly select any slice from repeating images ?
", "rawMarkdown": "Hi @nolialonso All the slice present in single images are of same category ? If this is the scenario we can randomly select any slice from repeating images ?\n", "isDeleted": true }, { "id": 2527868, "postDate": "2023-11-16T22:37:15.683Z", "content": "Technically you can choose a random slice, but it may be better to select the \"best\" slice, which would be clear, sharp, and have the least artifacts.
", "rawMarkdown": "Technically you can choose a random slice, but it may be better to select the \"best\" slice, which would be clear, sharp, and have the least artifacts.", "votes": 1 } ] } ] }, { "id": 2488547, "postDate": "2023-10-19T10:21:51.600Z", "content": "I saw them on Mayo Clinic - STRIP AI competition too. Why do they occur? Are they low quality versions of the tiff pyramidal images? Do they also exist on raw data or is it some kind of an artifact that happens while exporting the data to Kaggle?
", "rawMarkdown": "I saw them on Mayo Clinic - STRIP AI competition too. Why do they occur? Are they low quality versions of the tiff pyramidal images? Do they also exist on raw data or is it some kind of an artifact that happens while exporting the data to Kaggle?", "votes": 14 }, { "id": 2490369, "postDate": "2023-10-20T16:09:42.280Z", "content": "@nolialonso is correct, that's just how some labs prepare samples.
", "rawMarkdown": "@nolialonso is correct, that's just how some labs prepare samples.", "votes": 3 }, { "id": 2499402, "postDate": "2023-10-26T02:50:20.613Z", "content": "I investigated the repeating images and was able to detect those slides that had more than one image. Using contour I am able to crop these repeating images: https://www.kaggle.com/dhinkris/crop-duplicate-wsi-images
", "rawMarkdown": "I investigated the repeating images and was able to detect those slides that had more than one image. Using contour I am able to crop these repeating images: https://www.kaggle.com/dhinkris/crop-duplicate-wsi-images ", "votes": 4, "replies": [ { "id": 2527813, "postDate": "2023-11-16T20:28:37.637Z", "content": "@dhinkris really appreciate your work. What we need to do related to repeating images. Just take any random slice ?
", "rawMarkdown": "@dhinkris really appreciate your work. What we need to do related to repeating images. Just take any random slice ?", "isDeleted": true } ] }, { "id": 2525980, "postDate": "2023-11-15T14:13:54.117Z", "content": "Thanks for information. Do you mean that similar objects are in different pictures ? or similar objects in one picture ?
\nIf similar objects are in different pictures, we should be careful of validation to prevent data leakage, I think.
During histopathology tissue processing, very thin slices are cut from the wax tissue block. Since glass slides are rectangular, they can usually fit 2 slices from the block. If the tissue in the wax block in small and thin, they can fit even more slices.
\nThen probably depending on the slide scanner, scanning settings, and annotator, the saved image is either the whole glass slide, or the best representative slice.
\nIn the training dataset, I think the most in one slide is 6 slices.
\nI vaguely remember reading an article or study about whether to use all the slices vs just one slice for model training. I think they concluded that it's more cost effective or efficient to use one of the slices, since there are only subtle differences.
I thought they were artificial copies of each other. How can they cut slices that look almost identical?
", "votes": 0, "replies": [ { "id": 2489916, "author_name": "Noli Alonso", "author_url": "", "post_date": "2023-10-20T09:42:19.940000", "content": "Here is a link to how tissue processing works https://www.leicabiosystems.com/knowledge-pathway/an-introduction-to-specimen-processing/.
\nThey look the same since the cut slices are less than 2 micrometers thick.
", "votes": 7, "replies": [] }, { "id": 2527818, "author_name": "", "author_url": "", "post_date": "2023-11-16T20:30:58.050000", "content": "Hi @nolialonso All the slice present in single images are of same category ? If this is the scenario we can randomly select any slice from repeating images ?
", "votes": 0, "replies": [] }, { "id": 2527868, "author_name": "Noli Alonso", "author_url": "", "post_date": "2023-11-16T22:37:15.683000", "content": "Technically you can choose a random slice, but it may be better to select the \"best\" slice, which would be clear, sharp, and have the least artifacts.
", "votes": 1, "replies": [] } ] } ] }, { "id": 2490369, "author_name": "Sohier Dane", "author_url": "", "post_date": "2023-10-20T16:09:42.280000", "content": "@nolialonso is correct, that's just how some labs prepare samples.
", "votes": 3, "replies": [] }, { "id": 2499402, "author_name": "dhinesh", "author_url": "", "post_date": "2023-10-26T02:50:20.613000", "content": "I investigated the repeating images and was able to detect those slides that had more than one image. Using contour I am able to crop these repeating images: https://www.kaggle.com/dhinkris/crop-duplicate-wsi-images
", "votes": 4, "replies": [ { "id": 2527813, "author_name": "", "author_url": "", "post_date": "2023-11-16T20:28:37.637000", "content": "@dhinkris really appreciate your work. What we need to do related to repeating images. Just take any random slice ?
", "votes": 0, "replies": [] } ] }, { "id": 2525980, "author_name": "Aurora_blue", "author_url": "", "post_date": "2023-11-15T14:13:54.117000", "content": "Thanks for information. Do you mean that similar objects are in different pictures ? or similar objects in one picture ?
\nIf similar objects are in different pictures, we should be careful of validation to prevent data leakage, I think.