![\"\"](\"https://www.researchgate.net/publication/329715894/figure/fig1/AS:734044610383872@1552021281002/System-overview-An-illustration-of-the-explainable-deep-learning-system-for-ICH-detection.png\")
![\"\"](\"https://www.researchgate.net/publication/329715894/figure/fig2/AS:734044610392064@1552021281089/of-the-system-outputs-An-example-of-a-case-with-both-SAH-and-IVH-a-Probabilities-for.png\")
Hybrid 3D/2D Convolutional Neural Network for Hemorrhage Evaluation on Head CT
\n\n\nwas used to develop and cross-validate a custom hybrid 3D/2D mask ROI-based convolutional neural network architecture for hemorrhage evaluation.
\n
Brain Hemorrhage Diagnosis by Using Deep Learning
\n\n\nthe used: LeNet, GoogLeNet, and Inception-ResNet
\n
\nI'll keep updating this :)
Hybrid 3D/2D Convolutional Neural Network for Hemorrhage Evaluation on Head CT
\n\n\nwas used to develop and cross-validate a custom hybrid 3D/2D mask ROI-based convolutional neural network architecture for hemorrhage evaluation.
\n
Brain Hemorrhage Diagnosis by Using Deep Learning
\n\n\nthe used: LeNet, GoogLeNet, and Inception-ResNet
\n
\nI'll keep updating this :)
Nice! Excited to dive into the 3D neural net stuff! Thanks for sharing!
", "rawMarkdown": "Nice! Excited to dive into the 3D neural net stuff! Thanks for sharing!", "votes": 1 }, { "id": 632844, "postDate": "2019-09-24T06:16:46.120Z", "content": "Thanks! The full text for Brain Hemorrhage Diagnosis by Using Deep Learning is here:\nhttps://www.researchgate.net/publication/315853163_Brain_Hemorrhage_Diagnosis_by_Using_Deep_Learning
", "rawMarkdown": "Thanks! The full text for *Brain Hemorrhage Diagnosis by Using Deep Learning* is here:\nhttps://www.researchgate.net/publication/315853163_Brain_Hemorrhage_Diagnosis_by_Using_Deep_Learning" }, { "id": 630264, "postDate": "2019-09-20T02:16:46.123Z", "content": "The paper \"Brain Hemorrhage Diagnosis by Deep Learning\" used a pre-trained network? I'm not an expert in medical imaging, but I always got the impression that pre-trained networks don't work that well for very specialized tasks, like medical diagnosis?
", "rawMarkdown": "The paper \"Brain Hemorrhage Diagnosis by Deep Learning\" used a pre-trained network? I'm not an expert in medical imaging, but I always got the impression that pre-trained networks don't work that well for very specialized tasks, like medical diagnosis?", "replies": [ { "id": 632214, "postDate": "2019-09-23T11:18:22.197Z", "content": "I think we usually want to start with a pre-trained network if possible, because even if it's a very different task it's better than initialising the weights randomly.
", "rawMarkdown": "I think we usually want to start with a pre-trained network if possible, because even if it's a very different task it's better than initialising the weights randomly. ", "votes": 1 }, { "id": 636083, "postDate": "2019-09-28T19:42:06.353Z", "content": "I believe that the convolution layers at the very beginning of a pretrained network usually detect lines, edges, or some primitive shapes, etc. So even with this case, that is very different from ImageNet, it could be helpful to start with some pretrained filters.
\n\nFor example, in the recent protein competition, the convergence of the network with preloaded weights was much faster. Though you need to gradually un-freeze and fine-tune all your layers to adapt them for a new domain, of course.
", "rawMarkdown": "I believe that the convolution layers at the very beginning of a pretrained network usually detect lines, edges, or some primitive shapes, etc. So even with this case, that is very different from ImageNet, it could be helpful to start with some pretrained filters. \n\nFor example, in the recent protein competition, the convergence of the network with preloaded weights was much faster. Though you need to gradually un-freeze and fine-tune all your layers to adapt them for a new domain, of course." }, { "id": 636087, "postDate": "2019-09-28T19:49:38.803Z", "content": "@frankkloster transfer learning is the next big thing,it works extremely well with medical diagnosis datasets but for that you must use transfer learning techniques very wisely because it can easily be overfitted if you do not spend long enough time researching
", "rawMarkdown": "@frankkloster transfer learning is the next big thing,it works extremely well with medical diagnosis datasets but for that you must use transfer learning techniques very wisely because it can easily be overfitted if you do not spend long enough time researching" } ] }, { "id": 629533, "postDate": "2019-09-18T22:09:40.817Z", "content": "maybe i will try Inception-ResNet thanks @jesucristo
", "rawMarkdown": "maybe i will try Inception-ResNet thanks @jesucristo " }, { "id": 629997, "postDate": "2019-09-19T14:55:47.153Z", "content": "Thanks for this route of reading. 👍
", "rawMarkdown": "Thanks for this route of reading. 👍 " } ], "comments": [ { "id": 633155, "author_name": "Carlo Lepelaars", "author_url": "", "post_date": "2019-09-24T13:48:14.973000", "content": "Nice! Excited to dive into the 3D neural net stuff! Thanks for sharing!
", "votes": 1, "replies": [] }, { "id": 632844, "author_name": "Josh Myers", "author_url": "", "post_date": "2019-09-24T06:16:46.120000", "content": "Thanks! The full text for Brain Hemorrhage Diagnosis by Using Deep Learning is here:\nhttps://www.researchgate.net/publication/315853163_Brain_Hemorrhage_Diagnosis_by_Using_Deep_Learning
", "votes": 0, "replies": [] }, { "id": 630264, "author_name": "Frank Kloster", "author_url": "", "post_date": "2019-09-20T02:16:46.123000", "content": "The paper \"Brain Hemorrhage Diagnosis by Deep Learning\" used a pre-trained network? I'm not an expert in medical imaging, but I always got the impression that pre-trained networks don't work that well for very specialized tasks, like medical diagnosis?
", "votes": 0, "replies": [ { "id": 632214, "author_name": "Josh Myers", "author_url": "", "post_date": "2019-09-23T11:18:22.197000", "content": "I think we usually want to start with a pre-trained network if possible, because even if it's a very different task it's better than initialising the weights randomly.
", "votes": 1, "replies": [] }, { "id": 636083, "author_name": "Ilia Zaitsev", "author_url": "", "post_date": "2019-09-28T19:42:06.353000", "content": "I believe that the convolution layers at the very beginning of a pretrained network usually detect lines, edges, or some primitive shapes, etc. So even with this case, that is very different from ImageNet, it could be helpful to start with some pretrained filters.
\n\nFor example, in the recent protein competition, the convergence of the network with preloaded weights was much faster. Though you need to gradually un-freeze and fine-tune all your layers to adapt them for a new domain, of course.
", "votes": 0, "replies": [] }, { "id": 636087, "author_name": "Mobassir", "author_url": "", "post_date": "2019-09-28T19:49:38.803000", "content": "@frankkloster transfer learning is the next big thing,it works extremely well with medical diagnosis datasets but for that you must use transfer learning techniques very wisely because it can easily be overfitted if you do not spend long enough time researching
", "votes": 0, "replies": [] } ] }, { "id": 629533, "author_name": "Mobassir", "author_url": "", "post_date": "2019-09-18T22:09:40.817000", "content": "maybe i will try Inception-ResNet thanks @jesucristo
", "votes": 0, "replies": [] }, { "id": 629997, "author_name": "Marco Vasquez E", "author_url": "", "post_date": "2019-09-19T14:55:47.153000", "content": "Thanks for this route of reading. 👍
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "629514": "# Intracranial Hemorrhage Detection +Deep Learning\n\n\n\n\n\n