First in my mind for that task is the use of convolutional neural networks (CNNs) as a basis. I chose only the most relevant resources as I'm not a fan of overloading to avoid getting lost.
\nThis is a non-exhaustive list, I recommend to have a look at the previous competition of RSNA, especially 2022 & 2023. Feel free to add more in the comments for further sharing :)
", "messages": [ { "id": "2818148", "postDate": "05/17/2024 09:52:56", "content": "First in my mind for that task is the use of convolutional neural networks (CNNs) as a basis. I chose only the most relevant resources as I'm not a fan of overloading to avoid getting lost.
\nThis is a non-exhaustive list, I recommend to have a look at the previous competition of RSNA, especially 2022 & 2023. Feel free to add more in the comments for further sharing :)
", "rawMarkdown": "First in my mind for that task is the use of convolutional neural networks (CNNs) as a basis. I chose only the most relevant resources as I'm not a fan of overloading to avoid getting lost.\n\n## Domain Knowledge\n- I read the following [paper](https://www.researchgate.net/publication/376464794_LUMBAR_SPINE_DISEASE_DETECTION_Enhanced_CNN_Model_with_Improved_Classification_Accuracy) which has a lot useful insights for the domain in general and they used architecture.\n\"The high classification accuracy of 96% is achieved with multi support vector machine (MSVM), 94% with random forest (RF), 93.5% with a decision tree (DT), and 91% with the Naïve Bayes (NB) approach, proving the validity of the proposed approach.\"\n- Excellent explanation video of the Lumbar Degenerative [Part 1](https://www.youtube.com/watch?v=paSbo8We2oQ) by Dr. Brian Su - The Spine Guy. He also has various other videos on that topic. \n\n## Notebooks\n- [First place solution of RSNA 2022 Cervical Spine Fracture Detection](https://www.kaggle.com/code/haqishen/rsna-2022-1st-place-solution-train-stage2-type2) by @haqishen \n- [EDA of of RSNA 2022 Fracture Detection](https://www.kaggle.com/code/samuelcortinhas/rsna-fracture-detection-in-depth-eda) by @samuelcortinhas \n- [First place solition of RSNA 2023 Architecture/Discussion](https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/447449) by @nischaydnk \n- [Osteoarthritis Classification using MobileNetV3](https://www.kaggle.com/code/vencerlanz09/osteoarthritis-classification-using-mobilenetv3) by @vencerlanz09 \n\nThis is a non-exhaustive list, I recommend to have a look at the [previous competition of RSNA](https://www.kaggle.com/organizations/RSNA/competitions), especially 2022 & 2023. Feel free to add more in the comments for further sharing :)", "votes": null }, { "id": "2822962", "postDate": "05/18/2024 22:28:50", "content": "Thank you for sharing these resources! I was unsure about where to start my research for this competition, and your post has given me a clear direction. The paper on classification accuracy and the video by Dr. Brian Su seem particularly insightful. I appreciate the recommendations for notebooks as well.
\nThanks again!
", "rawMarkdown": "Thank you for sharing these resources! I was unsure about where to start my research for this competition, and your post has given me a clear direction. The paper on classification accuracy and the video by Dr. Brian Su seem particularly insightful. I appreciate the recommendations for notebooks as well. \n\nThanks again!", "votes": null }, { "id": "2823449", "postDate": "05/19/2024 07:42:43", "content": "I'm glad it's helping you! I agree with you; knowing where to start isn't always easy or straightforward.
", "rawMarkdown": "I'm glad it's helping you! I agree with you; knowing where to start isn't always easy or straightforward.", "votes": null }, { "id": "2904715", "postDate": "07/04/2024 14:34:34", "content": "Thanks for the resources @etiennekaiser. I have been going through the paper you mentioned for general knowledge; have you tried replicating their work for this challenge or do you know any githuub source code for verifying their work?
", "rawMarkdown": "Thanks for the resources @etiennekaiser. I have been going through the paper you mentioned for general knowledge; have you tried replicating their work for this challenge or do you know any githuub source code for verifying their work?", "votes": null }, { "id": "2905320", "postDate": "07/04/2024 20:23:00", "content": "Hello @rahulnakka - thanks for your feedback! Unfortunately, their code is not open source. However, I found that they tried quite a lot of models in an ensemble, and you can read about how well they performed. I would refer also to their architecture (Figure 4) and metrics visualization (Figure 10) for more details.
\nAs a starting point, they mainly used the following models as an ensemble:
\nAfter experimenting with these classical machine learning algorithms, you can also add some deep learning models to represent more complex patterns. They tried models such as CNN, RNN, VGG 16, ResNet50, and SE ResNet50. I didn't reproduce the results, but I would start with an MSVM as a strong base model.
", "rawMarkdown": "Hello @rahulnakka - thanks for your feedback! Unfortunately, their code is not open source. However, I found that they tried quite a lot of models in an ensemble, and you can read about how well they performed. I would refer also to their architecture (Figure 4) and metrics visualization (Figure 10) for more details.\n\nAs a starting point, they mainly used the following models as an ensemble:\n\n- Multi Support Vector Machine (MSVM) => 96% classification accuracy\n- Random Forest (RF) => 94% classification accuracy\n- Decision Tree (DT) => 93.5% classification accuracy\n- Naïve Bayes (NB) => 91% classification accuracy\n\nAfter experimenting with these classical machine learning algorithms, you can also add some deep learning models to represent more complex patterns. They tried models such as CNN, RNN, VGG 16, ResNet50, and SE ResNet50. I didn't reproduce the results, but I would start with an MSVM as a strong base model.", "votes": null } ], "comments": [ { "id": 2822962, "author_name": "mrnosaj", "author_url": "", "post_date": "05/18/2024 22:28:50", "content": "Thank you for sharing these resources! I was unsure about where to start my research for this competition, and your post has given me a clear direction. The paper on classification accuracy and the video by Dr. Brian Su seem particularly insightful. I appreciate the recommendations for notebooks as well.
\nThanks again!
", "votes": null, "replies": [ { "id": 2823449, "author_name": "etiennekaiser", "author_url": "", "post_date": "05/19/2024 07:42:43", "content": "I'm glad it's helping you! I agree with you; knowing where to start isn't always easy or straightforward.
", "votes": null, "replies": [] } ] }, { "id": 2904715, "author_name": "rahulnakka", "author_url": "", "post_date": "07/04/2024 14:34:34", "content": "Thanks for the resources @etiennekaiser. I have been going through the paper you mentioned for general knowledge; have you tried replicating their work for this challenge or do you know any githuub source code for verifying their work?
", "votes": null, "replies": [ { "id": 2905320, "author_name": "etiennekaiser", "author_url": "", "post_date": "07/04/2024 20:23:00", "content": "Hello @rahulnakka - thanks for your feedback! Unfortunately, their code is not open source. However, I found that they tried quite a lot of models in an ensemble, and you can read about how well they performed. I would refer also to their architecture (Figure 4) and metrics visualization (Figure 10) for more details.
\nAs a starting point, they mainly used the following models as an ensemble:
\nAfter experimenting with these classical machine learning algorithms, you can also add some deep learning models to represent more complex patterns. They tried models such as CNN, RNN, VGG 16, ResNet50, and SE ResNet50. I didn't reproduce the results, but I would start with an MSVM as a strong base model.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2818148": "First in my mind for that task is the use of convolutional neural networks (CNNs) as a basis. I chose only the most relevant resources as I'm not a fan of overloading to avoid getting lost.\n\n## Domain Knowledge\n- I read the following [paper](https://www.researchgate.net/publication/376464794_LUMBAR_SPINE_DISEASE_DETECTION_Enhanced_CNN_Model_with_Improved_Classification_Accuracy) which has a lot useful insights for the domain in general and they used architecture.\n\"The high classification accuracy of 96% is achieved with multi support vector machine (MSVM), 94% with random forest (RF), 93.5% with a decision tree (DT), and 91% with the Naïve Bayes (NB) approach, proving the validity of the proposed approach.\"\n- Excellent explanation video of the Lumbar Degenerative [Part 1](https://www.youtube.com/watch?v=paSbo8We2oQ) by Dr. Brian Su - The Spine Guy. He also has various other videos on that topic. \n\n## Notebooks\n- [First place solution of RSNA 2022 Cervical Spine Fracture Detection](https://www.kaggle.com/code/haqishen/rsna-2022-1st-place-solution-train-stage2-type2) by @haqishen \n- [EDA of of RSNA 2022 Fracture Detection](https://www.kaggle.com/code/samuelcortinhas/rsna-fracture-detection-in-depth-eda) by @samuelcortinhas \n- [First place solition of RSNA 2023 Architecture/Discussion](https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/447449) by @nischaydnk \n- [Osteoarthritis Classification using MobileNetV3](https://www.kaggle.com/code/vencerlanz09/osteoarthritis-classification-using-mobilenetv3) by @vencerlanz09 \n\nThis is a non-exhaustive list, I recommend to have a look at the [previous competition of RSNA](https://www.kaggle.com/organizations/RSNA/competitions), especially 2022 & 2023. Feel free to add more in the comments for further sharing :)", "2822962": "Thank you for sharing these resources! I was unsure about where to start my research for this competition, and your post has given me a clear direction. The paper on classification accuracy and the video by Dr. Brian Su seem particularly insightful. I appreciate the recommendations for notebooks as well. \n\nThanks again!", "2823449": "I'm glad it's helping you! I agree with you; knowing where to start isn't always easy or straightforward.", "2904715": "Thanks for the resources @etiennekaiser. I have been going through the paper you mentioned for general knowledge; have you tried replicating their work for this challenge or do you know any githuub source code for verifying their work?", "2905320": "Hello @rahulnakka - thanks for your feedback! Unfortunately, their code is not open source. However, I found that they tried quite a lot of models in an ensemble, and you can read about how well they performed. I would refer also to their architecture (Figure 4) and metrics visualization (Figure 10) for more details.\n\nAs a starting point, they mainly used the following models as an ensemble:\n\n- Multi Support Vector Machine (MSVM) => 96% classification accuracy\n- Random Forest (RF) => 94% classification accuracy\n- Decision Tree (DT) => 93.5% classification accuracy\n- Naïve Bayes (NB) => 91% classification accuracy\n\nAfter experimenting with these classical machine learning algorithms, you can also add some deep learning models to represent more complex patterns. They tried models such as CNN, RNN, VGG 16, ResNet50, and SE ResNet50. I didn't reproduce the results, but I would start with an MSVM as a strong base model." }, "source": "meta" }