{ "id": 239891, "title": "Recap top solutions for Xray Medical Imaging comps", "url": "/competitions/siim-covid19-detection/discussion/239891", "author_name": "", "post_date": "2021-05-18T01:21:14.703896Z", "votes": 219, "comment_count": 16, "views": 0, "content": "

This is an object detection and classification problem.

\n

VinBigData Chest X-ray Abnormalities Detection

\n

\"\"

\n\n

Components: Detection models (YOLO-V4 ), Specialized detector for aortic enlargement, Multi-label classifier-based post-processing. Image size 1280.

\n\n

RSNA Pneumonia Detection Challenge

\n\n

SIIM-ACR Pneumothorax Segmentation

\n\n

RSNA STR Pulmonary Embolism Detection

", "messages": [ { "id": "1312296", "postDate": "05/18/2021 01:21:14", "content": "

This is an object detection and classification problem.

\n

VinBigData Chest X-ray Abnormalities Detection

\n

\"\"

\n\n

Components: Detection models (YOLO-V4 ), Specialized detector for aortic enlargement, Multi-label classifier-based post-processing. Image size 1280.

\n\n

RSNA Pneumonia Detection Challenge

\n\n

SIIM-ACR Pneumothorax Segmentation

\n\n

RSNA STR Pulmonary Embolism Detection

", "rawMarkdown": "`This is an object detection and classification problem.`\n\n## [VinBigData Chest X-ray Abnormalities Detection](https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection/overview)\n\n![](https://i.ibb.co/8MqmxBz/results-16-0.png =300x*)\n\n- [1st place solution](https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection/discussion/231511) + [code](https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection/discussion/232393) by @morizin \n \n - Base models: Detectron2 Resnet101 , YoloV5, EffDetD2\n - Ensemble: WBF from @zfturbo's https://github.com/ZFTurbo/Weighted-Boxes-Fusion. \n - open code, thanks!\n\n- [2nd place solution](https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection/discussion/229740) by @ivanpan \n\n - Based on MMDetection: Cascade_RFP_R50 , GFL_R101 , GFL_X101 , RetinaNet_X101.\n - Base size 1024x1024 , training with FP16.\n - Albumentations like: ShiftScaleRotate, IAAAffine, Blur/GaussianBlur/MedianBlur, RandomBrightnessContrast, IAAAdditiveGaussianNoise/GaussNoise, HorizontalFlip.\n\n- [3rd place solution](https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection/discussion/229769) by @scusywxy \n\nComponents: Detection models (YOLO-V4 ), Specialized detector for aortic enlargement, Multi-label classifier-based post-processing. Image size 1280.\n\n- [4th place](https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection/discussion/229786) by @hiraiitsuki \n\n - model: yolov5x\n - image size: 640\n - TTA: 3 scale patterns and horizontal flip\n - ensemble: (4fold cv * 3 different preprocessed labels ) = 12 models\n\n\n## [RSNA Pneumonia Detection Challenge](https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/)\n\n- [1st place](https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/70421) + [code](https://github.com/i-pan/kaggle-rsna18) by @vaillant\n\n - classification-detection pipeline\n - detection: RetinaNet, Deformable R-FCN, Deformable Relation Networks.\n - classification: InceptionResNetV2 , Xception , DenseNet169.\n - Boxes were ensembled using: https://github.com/ahrnbom/ensemble-objdet\n\n- [2nd place](https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/70427) by @dmytropoplavskiy \n - base model: custom RetinaNet (se-resnext101)\n - 512x512 resolution\n - augmentations: Mild rotations (up to 6 deg), shift, scale, shear and h_flip, for some images random level of blur and noise and gamma changes.\n - ensemble: NMS\n\n- [3rd place](https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/70632) + [code](https://github.com/pmcheng/rsna-pneumonia) by @pmcheng \n\n - base models: RetinaNet (resnet-50 and resnet-101 ) + focal loss \n - 224 x 224 resolution as a abdominal radiologist he considered that `high image resolution was not necessary for pneumonia bounding box prediction.`\n - augmentations: rotation, translation, scaling, and horizontal flipping + random constants\n - NMS to eliminate any overlapping bounding boxes\n\n\n## [SIIM-ACR Pneumothorax Segmentation](https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation/overview)\n\n- [1st place](https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation/discussion/107824) + [code](https://github.com/sneddy/kaggle-pneumathorax) by @sneddy\n\n - base models: AlbuNet (resnet34) , Resnet50 , SCSEUnet (seresnext50) \n - Combo loss: combinations of BCE, dice and focal.\n - start with 512x512 and uptrain on size 1024x1024\n - small batch (2-4) size without accumulation\n - detailed tricks on his summary and github\n\n- [2nd place](https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation/discussion/108009) + [code](https://github.com/yelanlan/Pneumothorax-Segmentation-2nd-place-solution) by @lanjunyelan\n\n - classification and segmentation pipeline.\n - Classification: classify whether an image in related with pneumothorax or not. Multi-task model based on UNET (seresnext 50, seresnext101, efficientnet-b3\n) with a branch for classifying. BCE + focal loss. Basic augmentation: hflip, scale, rotate, bright, blur\n\n - Segmentation: 2 base models: unet and deeplabv3. Loss: dice loss. Augmentation: same as classification\n\n- [3rd place](https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation/discussion/107981) + [code](https://github.com/bestfitting/kaggle/tree/master/siim_acr) by @bestfitting\n\n - base model UNET (resnet34, se-resnext50)\n - cropped lungs, 576x576 cropped images (1024x1024 initially).\n - Attention: CBAM\n - Loss: Lovasz Los.\n - No classification model, No classification loss\n\n## [RSNA STR Pulmonary Embolism Detection](https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection)", "votes": null }, { "id": "1312339", "postDate": "05/18/2021 02:07:00", "content": "

A great resourc. Thank You

", "rawMarkdown": "A great resourc. Thank You", "votes": null }, { "id": "1315432", "postDate": "05/19/2021 19:28:25", "content": "

One step at a time. Don't Panic!

", "rawMarkdown": "One step at a time. Don't Panic!", "votes": null }, { "id": "1319207", "postDate": "05/23/2021 02:57:55", "content": "

Thanks for sharing @jesucristo

\n

Since this topic is already here I thought I would just post this here instead of starting my own discussion.

\n

Another similar competition was Data Science Bowl 2017

\n

Here are some solution posts I found:

\n", "rawMarkdown": "Thanks for sharing @jesucristo \n\nSince this topic is already here I thought I would just post this here instead of starting my own discussion.\n\nAnother similar competition was [Data Science Bowl 2017](https://www.kaggle.com/c/data-science-bowl-2017/)\n\nHere are some solution posts I found:\n\n- 2nd place: https://www.kaggle.com/c/data-science-bowl-2017/discussion/31551\n- 9th place: https://eliasvansteenkiste.github.io/machine%20learning/lung-cancer-pred/", "votes": null }, { "id": "1322025", "postDate": "05/25/2021 06:39:55", "content": "

Great solutions. Thank you

", "rawMarkdown": "Great solutions. Thank you", "votes": null }, { "id": "1326122", "postDate": "05/28/2021 08:28:18", "content": "

I was looking for this!! Thank you for sharing!!

", "rawMarkdown": "I was looking for this!! Thank you for sharing!!", "votes": null }, { "id": "1327789", "postDate": "05/29/2021 16:11:11", "content": "

Thanks for your solution！

", "rawMarkdown": "Thanks for your solution！", "votes": null }, { "id": "1329037", "postDate": "05/30/2021 19:36:24", "content": "

Hey, hey, the RSNA STR Pulmonary Embolism Detection top solutions:
\n1st: https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/discussion/194145
\n2nd: https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/discussion/193401
\n4th: https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/discussion/193970

", "rawMarkdown": "Hey, hey, the RSNA STR Pulmonary Embolism Detection top solutions:\n1st: https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/discussion/194145\n2nd: https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/discussion/193401\n4th: https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/discussion/193970", "votes": null }, { "id": "1330637", "postDate": "06/01/2021 03:09:11", "content": "

Thanks for the work, it really helps a lot.

", "rawMarkdown": "Thanks for the work, it really helps a lot.", "votes": null }, { "id": "1331443", "postDate": "06/01/2021 13:22:37", "content": "

big help， great job！！

", "rawMarkdown": "big help， great job！！", "votes": null }, { "id": "1332751", "postDate": "06/02/2021 09:21:32", "content": "

thank you.

", "rawMarkdown": "thank you.", "votes": null }, { "id": "1341090", "postDate": "06/08/2021 12:26:16", "content": "

Thanks for sharing!

", "rawMarkdown": "Thanks for sharing!", "votes": null }, { "id": "1343073", "postDate": "06/10/2021 02:30:39", "content": "

Thank you for sharing!

", "rawMarkdown": "Thank you for sharing!", "votes": null }, { "id": "1351123", "postDate": "06/16/2021 04:53:21", "content": "

Hello! I don't know if this is still a problem, but image level set is still bugged, and still retrieving very low scores. This is why many people are still using image study level. Is this still true? Or am I missing something? Thanks!

", "rawMarkdown": "Hello! I don't know if this is still a problem, but image level set is still bugged, and still retrieving very low scores. This is why many people are still using image study level. Is this still true? Or am I missing something? Thanks!", "votes": null }, { "id": "1387191", "postDate": "07/14/2021 03:26:46", "content": "

Helpful post..Thanks for this

", "rawMarkdown": "Helpful post..Thanks for this", "votes": null }, { "id": "1397772", "postDate": "07/23/2021 13:37:21", "content": "

Thanks for summarizing @jesucristo

", "rawMarkdown": "Thanks for summarizing @jesucristo", "votes": null }, { "id": "2207806", "postDate": "04/03/2023 16:31:53", "content": "

Thanks for sharing, its helpful!

", "rawMarkdown": "Thanks for sharing, its helpful!", "votes": null } ], "comments": [ { "id": 1312339, "author_name": "iammir", "author_url": "", "post_date": "05/18/2021 02:07:00", "content": "

A great resourc. Thank You

", "votes": null, "replies": [] }, { "id": 1315432, "author_name": "apthagowda", "author_url": "", "post_date": "05/19/2021 19:28:25", "content": "

One step at a time. Don't Panic!

", "votes": null, "replies": [] }, { "id": 1319207, "author_name": "illidan7", "author_url": "", "post_date": "05/23/2021 02:57:55", "content": "

Thanks for sharing @jesucristo

\n

Since this topic is already here I thought I would just post this here instead of starting my own discussion.

\n

Another similar competition was Data Science Bowl 2017

\n

Here are some solution posts I found:

\n", "votes": null, "replies": [] }, { "id": 1322025, "author_name": "nhannguyenle", "author_url": "", "post_date": "05/25/2021 06:39:55", "content": "

Great solutions. Thank you

", "votes": null, "replies": [] }, { "id": 1326122, "author_name": "soohyun", "author_url": "", "post_date": "05/28/2021 08:28:18", "content": "

I was looking for this!! Thank you for sharing!!

", "votes": null, "replies": [] }, { "id": 1327789, "author_name": "guanyazhang", "author_url": "", "post_date": "05/29/2021 16:11:11", "content": "

Thanks for your solution！

", "votes": null, "replies": [] }, { "id": 1329037, "author_name": "ricafernandes", "author_url": "", "post_date": "05/30/2021 19:36:24", "content": "

Hey, hey, the RSNA STR Pulmonary Embolism Detection top solutions:
\n1st: https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/discussion/194145
\n2nd: https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/discussion/193401
\n4th: https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/discussion/193970

", "votes": null, "replies": [] }, { "id": 1330637, "author_name": "chunchiangshen", "author_url": "", "post_date": "06/01/2021 03:09:11", "content": "

Thanks for the work, it really helps a lot.

", "votes": null, "replies": [] }, { "id": 1331443, "author_name": "jackxu081", "author_url": "", "post_date": "06/01/2021 13:22:37", "content": "

big help， great job！！

", "votes": null, "replies": [] }, { "id": 1332751, "author_name": "tensorchoko", "author_url": "", "post_date": "06/02/2021 09:21:32", "content": "

thank you.

", "votes": null, "replies": [] }, { "id": 1341090, "author_name": "albusyuxuanwang", "author_url": "", "post_date": "06/08/2021 12:26:16", "content": "

Thanks for sharing!

", "votes": null, "replies": [] }, { "id": 1343073, "author_name": "shangweichen", "author_url": "", "post_date": "06/10/2021 02:30:39", "content": "

Thank you for sharing!

", "votes": null, "replies": [] }, { "id": 1351123, "author_name": "andyjianzhou", "author_url": "", "post_date": "06/16/2021 04:53:21", "content": "

Hello! I don't know if this is still a problem, but image level set is still bugged, and still retrieving very low scores. This is why many people are still using image study level. Is this still true? Or am I missing something? Thanks!

", "votes": null, "replies": [] }, { "id": 1387191, "author_name": "mayankghogale", "author_url": "", "post_date": "07/14/2021 03:26:46", "content": "

Helpful post..Thanks for this

", "votes": null, "replies": [] }, { "id": 1397772, "author_name": "amritpal333", "author_url": "", "post_date": "07/23/2021 13:37:21", "content": "

Thanks for summarizing @jesucristo

", "votes": null, "replies": [] }, { "id": 2207806, "author_name": "marufnthewindows", "author_url": "", "post_date": "04/03/2023 16:31:53", "content": "

Thanks for sharing, its helpful!

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1312296": "`This is an object detection and classification problem.`\n\n## [VinBigData Chest X-ray Abnormalities Detection](https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection/overview)\n\n![](https://i.ibb.co/8MqmxBz/results-16-0.png =300x*)\n\n- [1st place solution](https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection/discussion/231511) + [code](https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection/discussion/232393) by @morizin \n \n - Base models: Detectron2 Resnet101 , YoloV5, EffDetD2\n - Ensemble: WBF from @zfturbo's https://github.com/ZFTurbo/Weighted-Boxes-Fusion. \n - open code, thanks!\n\n- [2nd place solution](https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection/discussion/229740) by @ivanpan \n\n - Based on MMDetection: Cascade_RFP_R50 , GFL_R101 , GFL_X101 , RetinaNet_X101.\n - Base size 1024x1024 , training with FP16.\n - Albumentations like: ShiftScaleRotate, IAAAffine, Blur/GaussianBlur/MedianBlur, RandomBrightnessContrast, IAAAdditiveGaussianNoise/GaussNoise, HorizontalFlip.\n\n- [3rd place solution](https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection/discussion/229769) by @scusywxy \n\nComponents: Detection models (YOLO-V4 ), Specialized detector for aortic enlargement, Multi-label classifier-based post-processing. Image size 1280.\n\n- [4th place](https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection/discussion/229786) by @hiraiitsuki \n\n - model: yolov5x\n - image size: 640\n - TTA: 3 scale patterns and horizontal flip\n - ensemble: (4fold cv * 3 different preprocessed labels ) = 12 models\n\n\n## [RSNA Pneumonia Detection Challenge](https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/)\n\n- [1st place](https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/70421) + [code](https://github.com/i-pan/kaggle-rsna18) by @vaillant\n\n - classification-detection pipeline\n - detection: RetinaNet, Deformable R-FCN, Deformable Relation Networks.\n - classification: InceptionResNetV2 , Xception , DenseNet169.\n - Boxes were ensembled using: https://github.com/ahrnbom/ensemble-objdet\n\n- [2nd place](https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/70427) by @dmytropoplavskiy \n - base model: custom RetinaNet (se-resnext101)\n - 512x512 resolution\n - augmentations: Mild rotations (up to 6 deg), shift, scale, shear and h_flip, for some images random level of blur and noise and gamma changes.\n - ensemble: NMS\n\n- [3rd place](https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/70632) + [code](https://github.com/pmcheng/rsna-pneumonia) by @pmcheng \n\n - base models: RetinaNet (resnet-50 and resnet-101 ) + focal loss \n - 224 x 224 resolution as a abdominal radiologist he considered that `high image resolution was not necessary for pneumonia bounding box prediction.`\n - augmentations: rotation, translation, scaling, and horizontal flipping + random constants\n - NMS to eliminate any overlapping bounding boxes\n\n\n## [SIIM-ACR Pneumothorax Segmentation](https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation/overview)\n\n- [1st place](https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation/discussion/107824) + [code](https://github.com/sneddy/kaggle-pneumathorax) by @sneddy\n\n - base models: AlbuNet (resnet34) , Resnet50 , SCSEUnet (seresnext50) \n - Combo loss: combinations of BCE, dice and focal.\n - start with 512x512 and uptrain on size 1024x1024\n - small batch (2-4) size without accumulation\n - detailed tricks on his summary and github\n\n- [2nd place](https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation/discussion/108009) + [code](https://github.com/yelanlan/Pneumothorax-Segmentation-2nd-place-solution) by @lanjunyelan\n\n - classification and segmentation pipeline.\n - Classification: classify whether an image in related with pneumothorax or not. Multi-task model based on UNET (seresnext 50, seresnext101, efficientnet-b3\n) with a branch for classifying. BCE + focal loss. Basic augmentation: hflip, scale, rotate, bright, blur\n\n - Segmentation: 2 base models: unet and deeplabv3. Loss: dice loss. Augmentation: same as classification\n\n- [3rd place](https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation/discussion/107981) + [code](https://github.com/bestfitting/kaggle/tree/master/siim_acr) by @bestfitting\n\n - base model UNET (resnet34, se-resnext50)\n - cropped lungs, 576x576 cropped images (1024x1024 initially).\n - Attention: CBAM\n - Loss: Lovasz Los.\n - No classification model, No classification loss\n\n## [RSNA STR Pulmonary Embolism Detection](https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection)", "1312339": "A great resourc. Thank You", "1315432": "One step at a time. Don't Panic!", "1319207": "Thanks for sharing @jesucristo \n\nSince this topic is already here I thought I would just post this here instead of starting my own discussion.\n\nAnother similar competition was [Data Science Bowl 2017](https://www.kaggle.com/c/data-science-bowl-2017/)\n\nHere are some solution posts I found:\n\n- 2nd place: https://www.kaggle.com/c/data-science-bowl-2017/discussion/31551\n- 9th place: https://eliasvansteenkiste.github.io/machine%20learning/lung-cancer-pred/", "1322025": "Great solutions. Thank you", "1326122": "I was looking for this!! Thank you for sharing!!", "1327789": "Thanks for your solution！", "1329037": "Hey, hey, the RSNA STR Pulmonary Embolism Detection top solutions:\n1st: https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/discussion/194145\n2nd: https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/discussion/193401\n4th: https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/discussion/193970", "1330637": "Thanks for the work, it really helps a lot.", "1331443": "big help， great job！！", "1332751": "thank you.", "1341090": "Thanks for sharing!", "1343073": "Thank you for sharing!", "1351123": "Hello! I don't know if this is still a problem, but image level set is still bugged, and still retrieving very low scores. This is why many people are still using image study level. Is this still true? Or am I missing something? Thanks!", "1387191": "Helpful post..Thanks for this", "1397772": "Thanks for summarizing @jesucristo", "2207806": "Thanks for sharing, its helpful!" }, "source": "meta" }