Here I want to share my previous experience in detecting abnormal breasts on mammogram images. As you know, in recent years deep learning has revolutionized various tasks like classification, segmentation, etc. Most of the time, traditional methods in image processing are underestimated due to the great success of CNNs.
\nIn this GitHub repository, I have implemented a traditional method based on Wavelet Energy Entropy (WEE). Actually, I have tried to replicate the method of another Article.
\nDespite having too much data for training deep CNNs or ViTs (there are other problems here like imbalanced data), in my opinion, the traditional methods such as Wavelet can help CNNs too! For instance, fusing the extracted features might help. I will keep updating my discoveries on this approach here.
\nEnjoy Kaggling :)
", "messages": [ { "id": 2051120, "postDate": "2022-12-01T08:12:40.013Z", "content": "Here I want to share my previous experience in detecting abnormal breasts on mammogram images. As you know, in recent years deep learning has revolutionized various tasks like classification, segmentation, etc. Most of the time, traditional methods in image processing are underestimated due to the great success of CNNs.
\nIn this GitHub repository, I have implemented a traditional method based on Wavelet Energy Entropy (WEE). Actually, I have tried to replicate the method of another Article.
\nDespite having too much data for training deep CNNs or ViTs (there are other problems here like imbalanced data), in my opinion, the traditional methods such as Wavelet can help CNNs too! For instance, fusing the extracted features might help. I will keep updating my discoveries on this approach here.
\nEnjoy Kaggling :)
", "rawMarkdown": "Here I want to **share my previous experience** in detecting abnormal breasts on mammogram images. As you know, in recent years deep learning has revolutionized various tasks like classification, segmentation, etc. Most of the time, traditional methods in image processing are underestimated due to the great success of CNNs.\n\nIn this [**GitHub repository**](https://github.com/MohammadDehghan/WEE_mammogram), I have implemented a traditional method based on **Wavelet Energy Entropy (WEE)**. Actually, I have tried to replicate the method of another Article. \n\nDespite having too much data for training deep CNNs or ViTs (there are other problems here like imbalanced data), in my opinion, the traditional methods such as Wavelet can help CNNs too! For instance, fusing the extracted features might help. I will keep updating my discoveries on this approach here.\n\nEnjoy Kaggling :)\n\n", "votes": 3 }, { "id": 2051146, "postDate": "2022-12-01T08:46:12.103Z", "content": "Sounds like interesting work! Given the data in the competition, how much of the image is blank space, even using just the part for ROI generation might be very useful!
", "rawMarkdown": "Sounds like interesting work! Given the data in the competition, how much of the image is blank space, even using just the part for ROI generation might be very useful!", "votes": 1, "replies": [ { "id": 2051159, "postDate": "2022-12-01T08:57:42.260Z", "content": "Thanks, @radek1 for your attention. It is definitely true, but to discover the impact, nothing beats conducting experiments.✌️
", "rawMarkdown": "Thanks, @radek1 for your attention. It is definitely true, but to discover the impact, nothing beats conducting experiments.✌️", "votes": 1 }, { "id": 2051170, "postDate": "2022-12-01T09:04:34.280Z", "content": "Absolutely!!! 🙂 I find the greatest dilemma on Kaggle competitions is: \"so many things to try, so few hours in a day\" 😄
\nCouldn't agree with you more! 🙂
", "rawMarkdown": "Absolutely!!! 🙂 I find the greatest dilemma on Kaggle competitions is: \"so many things to try, so few hours in a day\" 😄\n\nCouldn't agree with you more! 🙂", "votes": 1 } ] } ], "comments": [ { "id": 2051146, "author_name": "Radek Osmulski", "author_url": "", "post_date": "2022-12-01T08:46:12.103000", "content": "Sounds like interesting work! Given the data in the competition, how much of the image is blank space, even using just the part for ROI generation might be very useful!
", "votes": 1, "replies": [ { "id": 2051159, "author_name": "Mohammad Dehghanmanshadi", "author_url": "", "post_date": "2022-12-01T08:57:42.260000", "content": "Thanks, @radek1 for your attention. It is definitely true, but to discover the impact, nothing beats conducting experiments.✌️
", "votes": 1, "replies": [] }, { "id": 2051170, "author_name": "Radek Osmulski", "author_url": "", "post_date": "2022-12-01T09:04:34.280000", "content": "Absolutely!!! 🙂 I find the greatest dilemma on Kaggle competitions is: \"so many things to try, so few hours in a day\" 😄
\nCouldn't agree with you more! 🙂
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "2051120": "Here I want to **share my previous experience** in detecting abnormal breasts on mammogram images. As you know, in recent years deep learning has revolutionized various tasks like classification, segmentation, etc. Most of the time, traditional methods in image processing are underestimated due to the great success of CNNs.\n\nIn this [**GitHub repository**](https://github.com/MohammadDehghan/WEE_mammogram), I have implemented a traditional method based on **Wavelet Energy Entropy (WEE)**. Actually, I have tried to replicate the method of another Article. \n\nDespite having too much data for training deep CNNs or ViTs (there are other problems here like imbalanced data), in my opinion, the traditional methods such as Wavelet can help CNNs too! For instance, fusing the extracted features might help. I will keep updating my discoveries on this approach here.\n\nEnjoy Kaggling :)\n\n", "2051146": "Sounds like interesting work! Given the data in the competition, how much of the image is blank space, even using just the part for ROI generation might be very useful!" } }