![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F83267%2Fb7e508e002dd691f0acf45fad9b88a40%2Ffix.png?generation=1669730891843894&alt=media\")
Hey!
\nReally excited about this competition! 🙂
\nTo get us up and running quickly I put together this EDA:
\n📊 EDA + training a fast.ai model + submission 🚀
\nIn it, I cover:
\nHope you find it useful! 🙂 If you do, would appreciate an upvote! 🙌 Best of luck with the challenge!
", "messages": [ { "id": 2047709, "postDate": "2022-11-29T01:23:50.157Z", "content": "Hey!
\nReally excited about this competition! 🙂
\nTo get us up and running quickly I put together this EDA:
\n📊 EDA + training a fast.ai model + submission 🚀
\nIn it, I cover:
\nHope you find it useful! 🙂 If you do, would appreciate an upvote! 🙌 Best of luck with the challenge!
", "rawMarkdown": "Hey!\n\nReally excited about this competition! 🙂\n\nTo get us up and running quickly I put together this EDA:\n\n[📊 EDA + training a fast.ai model + submission 🚀](https://www.kaggle.com/code/radek1/initial-eda-a-first-look-at-the-data)\n\nIn it, I cover:\n\n1. How the data is organized.\n2. What data do we have available (a quick glance at the metadata, overview of the DICOM format).\n3. The competition metric.\n4. Label distribution in the train set.\n5. Training a fast.ai model\n6. Making a submission.\n\nHope you find it useful! 🙂 If you do, would appreciate an upvote! 🙌 Best of luck with the challenge!", "votes": 13 }, { "id": 2048485, "postDate": "2022-11-29T13:29:37.367Z", "content": "Nice work @radek1! Keep in mind that mammography images are usually MONOCHROME1, which means the pixel intensity is inverted. Add something like this before displaying the images with imshow to make them display properly .. with denser tissue appearing brighter.
\npixels = np.max(image.pixel_array) - image.pixel_array
Thank you, @davidbroberts! Fixed this now.
\n
Also, will keep track of PhotometricInterpretation on dicom files going forward, and transform the image appropriately if need be 🙂
Thank you for point this out to me!
", "rawMarkdown": "Thank you, @davidbroberts! Fixed this now.\n\n\n\nAlso, will keep track of `PhotometricInterpretation` on dicom files going forward, and transform the image appropriately if need be 🙂\n\nThank you for point this out to me!" } ] }, { "id": 2084357, "postDate": "2023-01-03T12:39:37.360Z", "content": "Very helpful. Thank you @radek1.
\nI see you've focussed on the cancer class. A Brief Intro to Mammography suggests \"you cannot diagnose cancer on a screening mammogram\". Therefore, shouldn't the class of interest be the BI-RADS score as indicated in \"Our Task\"?
", "rawMarkdown": "Very helpful. Thank you @radek1.\n\nI see you've focussed on the cancer class. [A Brief Intro to Mammography](https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/369262url) suggests \"you cannot diagnose cancer on a screening mammogram\". Therefore, shouldn't the class of interest be the BI-RADS score as indicated in \"Our Task\"? " } ], "comments": [ { "id": 2048485, "author_name": "David Roberts", "author_url": "", "post_date": "2022-11-29T13:29:37.367000", "content": "Nice work @radek1! Keep in mind that mammography images are usually MONOCHROME1, which means the pixel intensity is inverted. Add something like this before displaying the images with imshow to make them display properly .. with denser tissue appearing brighter.
\npixels = np.max(image.pixel_array) - image.pixel_array
Thank you, @davidbroberts! Fixed this now.
\n
Also, will keep track of PhotometricInterpretation on dicom files going forward, and transform the image appropriately if need be 🙂
Thank you for point this out to me!
", "votes": 0, "replies": [] } ] }, { "id": 2084357, "author_name": "Julian Macnamara", "author_url": "", "post_date": "2023-01-03T12:39:37.360000", "content": "Very helpful. Thank you @radek1.
\nI see you've focussed on the cancer class. A Brief Intro to Mammography suggests \"you cannot diagnose cancer on a screening mammogram\". Therefore, shouldn't the class of interest be the BI-RADS score as indicated in \"Our Task\"?
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2047709": "Hey!\n\nReally excited about this competition! 🙂\n\nTo get us up and running quickly I put together this EDA:\n\n[📊 EDA + training a fast.ai model + submission 🚀](https://www.kaggle.com/code/radek1/initial-eda-a-first-look-at-the-data)\n\nIn it, I cover:\n\n1. How the data is organized.\n2. What data do we have available (a quick glance at the metadata, overview of the DICOM format).\n3. The competition metric.\n4. Label distribution in the train set.\n5. Training a fast.ai model\n6. Making a submission.\n\nHope you find it useful! 🙂 If you do, would appreciate an upvote! 🙌 Best of luck with the challenge!", "2048485": "Nice work @radek1! Keep in mind that mammography images are usually MONOCHROME1, which means the pixel intensity is inverted. Add something like this before displaying the images with imshow to make them display properly .. with denser tissue appearing brighter. \n\n`pixels = np.max(image.pixel_array) - image.pixel_array`", "2084357": "Very helpful. Thank you @radek1.\n\nI see you've focussed on the cancer class. [A Brief Intro to Mammography](https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/369262url) suggests \"you cannot diagnose cancer on a screening mammogram\". Therefore, shouldn't the class of interest be the BI-RADS score as indicated in \"Our Task\"? " } }