Recently I am focusing on tiles visulization, and figure out how to classify different ovarian subtypes. And I try to extract some feature from different subtypes. From my perspective, that's a good point to build a feature engineering.
\nI will list some useful resources I read, and help beginners to understand how to classify ovarain.
\nlink: https://news.ipathology.cn/article/4832.html
\nThis article will teach you how to classify HGSC, EC, CC and MC. Also, it contains some images with arrows which can give you a good understanding. Finally, this article will show a table, which list some good features to classify different ovarian subtypes.
\nlink: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6412907/
\nFor quick understanding, you just need to jump to section 3. I believe those pic and description about HGSC can help you better understand.
link: https://pubmed.ncbi.nlm.nih.gov/35204549/
\nFor quick understanding, you just need to scan figure 1.
\nAlso, I summary some good featurs on my notebook, LGSC tiles are similar with HGSC tiles. They are all composed by papillaes. But most of LGSC tiles contains a large number of occasional cells with larger nuclei.(You can find some dark red spots from tile images), We can find that most of tiles can show a large number of occasional cells with larger nuclei. That's a good criterion to extract useful tiles or patch.
One interesting latest convolution structure may be useful for this competion:
\nIn this paper, the author use a new type of deformable conv to better extract local tubular structures. In this competion, most of WSLs shows complex papillary fronds structure, I think that can be a good feature to help model classify the ovarian subtype.
\nDynamic Snake Convolution based on Topological Geometric Constraints for Tubular Structure Segmentation: https://arxiv.org/abs/2307.08388
\ncode: https://github.com/YaoleiQi/DSCNet/tree/main
\nDSC_CONV_structure: https://github.com/YaoleiQi/DSCNet/blob/main/DSCNet_2D_opensource/Code/DRIVE/DSCNet/S3_DSConv.py
If someone is interested about it, you can try to use this idea.(Since I am busy on EDA and feat engineering recently, I may try it in the future)
\nThis is what you are looking for(ππππππ):
\nThis is a model can classify different subtype of ovarian cancer! But it only support WSIs images. I belive someone can get higher accuracy improvement from this paper.
\nClassification of Epithelial Ovarian Carcinoma Whole-Slide Pathology Images Using Deep Transfer Learning: https://openreview.net/forum?id=VXdQD8B307
\ncode: https://github.com/AIMLab-UBC/MIDL2020
For more professional discussion, click link1.
\nFor more paper or resources to read, click link2,3.
I am new to kaggle, and I don't have a professional bg on medical. So I tried to filter some good resources for you guys, if you find any wrong views on my above resouces, welcome to point it out.
", "messages": [ { "id": 2522006, "postDate": "2023-11-12T09:28:30.483Z", "content": "Recently I am focusing on tiles visulization, and figure out how to classify different ovarian subtypes. And I try to extract some feature from different subtypes. From my perspective, that's a good point to build a feature engineering.
\nI will list some useful resources I read, and help beginners to understand how to classify ovarain.
\nlink: https://news.ipathology.cn/article/4832.html
\nThis article will teach you how to classify HGSC, EC, CC and MC. Also, it contains some images with arrows which can give you a good understanding. Finally, this article will show a table, which list some good features to classify different ovarian subtypes.
\nlink: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6412907/
\nFor quick understanding, you just need to jump to section 3. I believe those pic and description about HGSC can help you better understand.
link: https://pubmed.ncbi.nlm.nih.gov/35204549/
\nFor quick understanding, you just need to scan figure 1.
\nAlso, I summary some good featurs on my notebook, LGSC tiles are similar with HGSC tiles. They are all composed by papillaes. But most of LGSC tiles contains a large number of occasional cells with larger nuclei.(You can find some dark red spots from tile images), We can find that most of tiles can show a large number of occasional cells with larger nuclei. That's a good criterion to extract useful tiles or patch.
One interesting latest convolution structure may be useful for this competion:
\nIn this paper, the author use a new type of deformable conv to better extract local tubular structures. In this competion, most of WSLs shows complex papillary fronds structure, I think that can be a good feature to help model classify the ovarian subtype.
\nDynamic Snake Convolution based on Topological Geometric Constraints for Tubular Structure Segmentation: https://arxiv.org/abs/2307.08388
\ncode: https://github.com/YaoleiQi/DSCNet/tree/main
\nDSC_CONV_structure: https://github.com/YaoleiQi/DSCNet/blob/main/DSCNet_2D_opensource/Code/DRIVE/DSCNet/S3_DSConv.py
If someone is interested about it, you can try to use this idea.(Since I am busy on EDA and feat engineering recently, I may try it in the future)
\nThis is what you are looking for(ππππππ):
\nThis is a model can classify different subtype of ovarian cancer! But it only support WSIs images. I belive someone can get higher accuracy improvement from this paper.
\nClassification of Epithelial Ovarian Carcinoma Whole-Slide Pathology Images Using Deep Transfer Learning: https://openreview.net/forum?id=VXdQD8B307
\ncode: https://github.com/AIMLab-UBC/MIDL2020
For more professional discussion, click link1.
\nFor more paper or resources to read, click link2,3.
I am new to kaggle, and I don't have a professional bg on medical. So I tried to filter some good resources for you guys, if you find any wrong views on my above resouces, welcome to point it out.
", "rawMarkdown": "Recently I am focusing on tiles visulization, and figure out how to classify different ovarian subtypes. And I try to extract some feature from different subtypes. From my perspective, that's a good point to build a feature engineering.\n\nI will list some useful resources I read, and help beginners to understand how to classify ovarain.\n## 5 ovarian subtypes in this competition\n* HGSC(High-Grade Serous Carcinoma)\n* LGSC(Low-Grade Serous Carcinoma)\n* EC(Endometrioid Carcinoma)\n* CC(Clear Cell)\n* MC(Mucinous Carcinoma) \n\n## The most valuable article I read(πππππ)\nlink: [https://news.ipathology.cn/article/4832.html](url)\n\nThis article will teach you how to classify HGSC, EC, CC and MC. Also, it contains some images with arrows which can give you a good understanding. Finally, this article will show a table, which list some good features to classify different ovarian subtypes.\n\n## HGSC paper(πππ)\nlink: [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6412907/ ](url)\nFor quick understanding, you just need to jump to section 3. I believe those pic and description about HGSC can help you better understand.\n\n## LGSC paper(ππππ)\nlink: [https://pubmed.ncbi.nlm.nih.gov/35204549/](url)\nFor quick understanding, you just need to scan figure 1.\nAlso, I summary some good featurs on my notebook, LGSC tiles are similar with HGSC tiles. They are all composed by papillaes. But most of LGSC tiles contains a large number of occasional cells with larger nuclei.(You can find some dark red spots from tile images), We can find that most of tiles can show a large number of occasional cells with larger nuclei. That's a good criterion to extract useful tiles or patch.\n\n### DL paper(ππ)\nOne interesting latest convolution structure may be useful for this competion:\nIn this paper, the author use a new type of deformable conv to better extract local tubular structures. In this competion, most of WSLs shows complex papillary fronds structure, I think that can be a good feature to help model classify the ovarian subtype.\n**Dynamic Snake Convolution based on Topological Geometric Constraints for Tubular Structure Segmentation**: https://arxiv.org/abs/2307.08388\ncode: https://github.com/YaoleiQi/DSCNet/tree/main \nDSC_CONV_structure: https://github.com/YaoleiQi/DSCNet/blob/main/DSCNet_2D_opensource/Code/DRIVE/DSCNet/S3_DSConv.py\n\nIf someone is interested about it, you can try to use this idea.(Since I am busy on EDA and feat engineering recently, I may try it in the future)\n\nThis is what you are looking for(ππππππ):\nThis is a model can classify different subtype of ovarian cancer! But it only support WSIs images. I belive someone can get higher accuracy improvement from this paper. \n**Classification of Epithelial Ovarian Carcinoma Whole-Slide Pathology Images Using Deep Transfer Learning**: https://openreview.net/forum?id=VXdQD8B307\ncode: https://github.com/AIMLab-UBC/MIDL2020\n\n## Other\nFor more professional discussion, click link1.\nFor more paper or resources to read, click link2,3. \n1. https://www.kaggle.com/competitions/UBC-OCEAN/discussion/445804 \n2. https://www.kaggle.com/competitions/UBC-OCEAN/discussion/445470\n3. https://www.kaggle.com/competitions/UBC-OCEAN/discussion/445472\n\nI am new to kaggle, and I don't have a professional bg on medical. So I tried to filter some good resources for you guys, if you find any wrong views on my above resouces, welcome to point it out.", "votes": 14 } ], "comments": [], "raw_markdown_by_id": { "2522006": "Recently I am focusing on tiles visulization, and figure out how to classify different ovarian subtypes. And I try to extract some feature from different subtypes. From my perspective, that's a good point to build a feature engineering.\n\nI will list some useful resources I read, and help beginners to understand how to classify ovarain.\n## 5 ovarian subtypes in this competition\n* HGSC(High-Grade Serous Carcinoma)\n* LGSC(Low-Grade Serous Carcinoma)\n* EC(Endometrioid Carcinoma)\n* CC(Clear Cell)\n* MC(Mucinous Carcinoma) \n\n## The most valuable article I read(πππππ)\nlink: [https://news.ipathology.cn/article/4832.html](url)\n\nThis article will teach you how to classify HGSC, EC, CC and MC. Also, it contains some images with arrows which can give you a good understanding. Finally, this article will show a table, which list some good features to classify different ovarian subtypes.\n\n## HGSC paper(πππ)\nlink: [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6412907/ ](url)\nFor quick understanding, you just need to jump to section 3. I believe those pic and description about HGSC can help you better understand.\n\n## LGSC paper(ππππ)\nlink: [https://pubmed.ncbi.nlm.nih.gov/35204549/](url)\nFor quick understanding, you just need to scan figure 1.\nAlso, I summary some good featurs on my notebook, LGSC tiles are similar with HGSC tiles. They are all composed by papillaes. But most of LGSC tiles contains a large number of occasional cells with larger nuclei.(You can find some dark red spots from tile images), We can find that most of tiles can show a large number of occasional cells with larger nuclei. That's a good criterion to extract useful tiles or patch.\n\n### DL paper(ππ)\nOne interesting latest convolution structure may be useful for this competion:\nIn this paper, the author use a new type of deformable conv to better extract local tubular structures. In this competion, most of WSLs shows complex papillary fronds structure, I think that can be a good feature to help model classify the ovarian subtype.\n**Dynamic Snake Convolution based on Topological Geometric Constraints for Tubular Structure Segmentation**: https://arxiv.org/abs/2307.08388\ncode: https://github.com/YaoleiQi/DSCNet/tree/main \nDSC_CONV_structure: https://github.com/YaoleiQi/DSCNet/blob/main/DSCNet_2D_opensource/Code/DRIVE/DSCNet/S3_DSConv.py\n\nIf someone is interested about it, you can try to use this idea.(Since I am busy on EDA and feat engineering recently, I may try it in the future)\n\nThis is what you are looking for(ππππππ):\nThis is a model can classify different subtype of ovarian cancer! But it only support WSIs images. I belive someone can get higher accuracy improvement from this paper. \n**Classification of Epithelial Ovarian Carcinoma Whole-Slide Pathology Images Using Deep Transfer Learning**: https://openreview.net/forum?id=VXdQD8B307\ncode: https://github.com/AIMLab-UBC/MIDL2020\n\n## Other\nFor more professional discussion, click link1.\nFor more paper or resources to read, click link2,3. \n1. https://www.kaggle.com/competitions/UBC-OCEAN/discussion/445804 \n2. https://www.kaggle.com/competitions/UBC-OCEAN/discussion/445470\n3. https://www.kaggle.com/competitions/UBC-OCEAN/discussion/445472\n\nI am new to kaggle, and I don't have a professional bg on medical. So I tried to filter some good resources for you guys, if you find any wrong views on my above resouces, welcome to point it out." } }