I was wondering if there is any more information about the resolution of the images that could be made public.
\nRight now, it's statated that the slides are scanned at magnifications of 40X for TMA's and 20X for biopsies respectively, I was wondering if there is any information regarding how many microns-per-pixel (MPP) this would be.
\nI'd say this is important to know, because for a given magnification, the MPP can still be different. As a result, the actual scale of the image can also vary, which can throw off models like CNN's, since they are not scale invariant.
\nIf possible, please let us know
", "messages": [ { "id": 2505227, "postDate": "2023-10-30T13:21:55.513Z", "content": "I was wondering if there is any more information about the resolution of the images that could be made public.
\nRight now, it's statated that the slides are scanned at magnifications of 40X for TMA's and 20X for biopsies respectively, I was wondering if there is any information regarding how many microns-per-pixel (MPP) this would be.
\nI'd say this is important to know, because for a given magnification, the MPP can still be different. As a result, the actual scale of the image can also vary, which can throw off models like CNN's, since they are not scale invariant.
\nIf possible, please let us know
", "rawMarkdown": "I was wondering if there is any more information about the resolution of the images that could be made public.\n\nRight now, it's statated that the slides are scanned at magnifications of 40X for TMA's and 20X for biopsies respectively, I was wondering if there is any information regarding how many microns-per-pixel (MPP) this would be. \n\nI'd say this is important to know, because for a given magnification, the MPP can still be different. As a result, the actual scale of the image can also vary, which can throw off models like CNN's, since they are not scale invariant. \n\nIf possible, please let us know", "votes": 4 }, { "id": 2506253, "postDate": "2023-10-31T07:24:27.167Z", "content": "It would be good to know that information but it's not that crucial. Even if they share it for the training set, test set might have different pixel spacing values since some of the images are coming from different sources. Augmenting images accordingly is a more robust approach.
", "rawMarkdown": "It would be good to know that information but it's not that crucial. Even if they share it for the training set, test set might have different pixel spacing values since some of the images are coming from different sources. Augmenting images accordingly is a more robust approach.", "votes": 2, "replies": [ { "id": 2506586, "postDate": "2023-10-31T12:03:17.483Z", "content": "I will definitely agree with a strong augmentation routine to increase robustness.
\nIn terms variations from different sources, it would depend on the logistics.
\nOne option would be to send the actual slides to one location, where they are all scanned on the same scanner with the same parameters.
\nThe other option is that each hospital sends digital slides made on their scanners and settings, which would mean a wider variation in data and possibly mpp.
It would be good to know that information but it's not that crucial. Even if they share it for the training set, test set might have different pixel spacing values since some of the images are coming from different sources. Augmenting images accordingly is a more robust approach.
", "votes": 2, "replies": [ { "id": 2506586, "author_name": "Stephan", "author_url": "", "post_date": "2023-10-31T12:03:17.483000", "content": "I will definitely agree with a strong augmentation routine to increase robustness.
\nIn terms variations from different sources, it would depend on the logistics.
\nOne option would be to send the actual slides to one location, where they are all scanned on the same scanner with the same parameters.
\nThe other option is that each hospital sends digital slides made on their scanners and settings, which would mean a wider variation in data and possibly mpp.