{ "id": 505499, "title": "Dicoms, DL & Medical Imaging: OsiriX, Pydicom, Oro, Mango viewer. Dicom on Kaggle.", "url": "/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/505499", "author_name": "", "post_date": "2024-05-17T19:05:46.133881200Z", "votes": 49, "comment_count": 2, "views": 0, "content": "

Working with Dicoms

\n

OsiriX Dicom Viewer

\n

Pydicom

\n

Pydicom documentation

\n

Oro DicomA package for working with images in R.

\n

Mango: A viewer for medical research images.

\n

Kaggle Inspiration:

\n

Data Science Bowl 2017 In this Competition dataset, you are given over a thousand low-dose CT images from high-risk patients in DICOM format.

\n

DON'T see like a radiologist! -fastai By Jeremy Howard 5y ago

\n

Basic EDA + Data Visualization By Marco Vasquez E. - 5y ago

\n

Standardizing Unusual Dicoms By Hui Ming Lin

\n

Deep Learning for MRI

\n

By Jason A. Polzin

\n

Steps:

\n

\"Image quality inspection: In the first step, the author checked if the given localizer image is suitable to identify the plane for the desired anatomy. This is achieved by using a fiver layer, dyadic reduction regular CNN classification network model (that we call “LocalizerIQ-Net”) to identify slices with relevant anatomy, slices with artifacts and irrelevant slices. If the localizer image is not suitable for ISP, relevant feedback is provided to the scan operator. They used the built-in TensorFlow functions for image manipulation to achieve data augmentation during the training of LocalizerIQ-Net.\"

\n

\"Identification of anatomy coverage: Next, the author located the spatial-extent of the desired anatomy in the localizer images by incorporating a shape-based semantic image segmentation U-Net DL model (called “Coverage-Net”). This helps the next processing steps to be robust to changes in imaging parameter settings across hospitals and clinics as well as changes in shapes and sizes of the patient’s anatomy.\"

\n

\"Identification of precise plane orientation and location: For each desired anatomic structure, the author found the scan-plane that is best suited to image that structure using one or more image segmentation 3D U-Net models (called “Orientation-Net”). Orientation-Net directly segments the desired-planes on localizer images, which is then used to compute the orientation and location.\"

\n

\"The entire process took ~3.5 seconds on a high-performance CPU. The training and testing data for these DL models came from more than 1300 subjects and were acquired using various GE scanner models and field strengths from several clinical sites.\"

\n

https://blog.tensorflow.org/2019/03/intelligent-scanning-using-deep-learning.html

\n

Building Neural Network for Medical Imaging using Deep Learning in Tensorflow

\n

By Chandan Verma (Kaggler)

\n

Steps below were taken in order to make the data consumable by the CNN:

\n

Loading the Dicom Image, Converting the Pixel values to Hounsfield(HU) units,
\nResampling, ROI Generation, Normalization, Zero Centering, Resizing the data
\nSplitting the data and label and Creating the Convolutional neural network.\"

\n

Converting Pixel values to Hounsfield units(HU

\n

\"Converting Pixel values to Hounsfield units(HU): This helps us in segregating different parts of the body. Converting to Hounsfield units enable us to extract the lungs which is the area of interest to us in this case. More on Hounsfield units(HU) here. Below is the formula to compute HU from pixel values.\"

\n

hu = pixel_value * slope + intercept

\n

https://medium.com/@verma.chandan/building-neural-network-for-medical-imaging-using-deep-learning-in-tensorflow-part-1-ab993b7fb04f

", "messages": [ { "id": "2820940", "postDate": "05/17/2024 19:05:46", "content": "

Working with Dicoms

\n

OsiriX Dicom Viewer

\n

Pydicom

\n

Pydicom documentation

\n

Oro DicomA package for working with images in R.

\n

Mango: A viewer for medical research images.

\n

Kaggle Inspiration:

\n

Data Science Bowl 2017 In this Competition dataset, you are given over a thousand low-dose CT images from high-risk patients in DICOM format.

\n

DON'T see like a radiologist! -fastai By Jeremy Howard 5y ago

\n

Basic EDA + Data Visualization By Marco Vasquez E. - 5y ago

\n

Standardizing Unusual Dicoms By Hui Ming Lin

\n

Deep Learning for MRI

\n

By Jason A. Polzin

\n

Steps:

\n

\"Image quality inspection: In the first step, the author checked if the given localizer image is suitable to identify the plane for the desired anatomy. This is achieved by using a fiver layer, dyadic reduction regular CNN classification network model (that we call “LocalizerIQ-Net”) to identify slices with relevant anatomy, slices with artifacts and irrelevant slices. If the localizer image is not suitable for ISP, relevant feedback is provided to the scan operator. They used the built-in TensorFlow functions for image manipulation to achieve data augmentation during the training of LocalizerIQ-Net.\"

\n

\"Identification of anatomy coverage: Next, the author located the spatial-extent of the desired anatomy in the localizer images by incorporating a shape-based semantic image segmentation U-Net DL model (called “Coverage-Net”). This helps the next processing steps to be robust to changes in imaging parameter settings across hospitals and clinics as well as changes in shapes and sizes of the patient’s anatomy.\"

\n

\"Identification of precise plane orientation and location: For each desired anatomic structure, the author found the scan-plane that is best suited to image that structure using one or more image segmentation 3D U-Net models (called “Orientation-Net”). Orientation-Net directly segments the desired-planes on localizer images, which is then used to compute the orientation and location.\"

\n

\"The entire process took ~3.5 seconds on a high-performance CPU. The training and testing data for these DL models came from more than 1300 subjects and were acquired using various GE scanner models and field strengths from several clinical sites.\"

\n

https://blog.tensorflow.org/2019/03/intelligent-scanning-using-deep-learning.html

\n

Building Neural Network for Medical Imaging using Deep Learning in Tensorflow

\n

By Chandan Verma (Kaggler)

\n

Steps below were taken in order to make the data consumable by the CNN:

\n

Loading the Dicom Image, Converting the Pixel values to Hounsfield(HU) units,
\nResampling, ROI Generation, Normalization, Zero Centering, Resizing the data
\nSplitting the data and label and Creating the Convolutional neural network.\"

\n

Converting Pixel values to Hounsfield units(HU

\n

\"Converting Pixel values to Hounsfield units(HU): This helps us in segregating different parts of the body. Converting to Hounsfield units enable us to extract the lungs which is the area of interest to us in this case. More on Hounsfield units(HU) here. Below is the formula to compute HU from pixel values.\"

\n

hu = pixel_value * slope + intercept

\n

https://medium.com/@verma.chandan/building-neural-network-for-medical-imaging-using-deep-learning-in-tensorflow-part-1-ab993b7fb04f

", "rawMarkdown": "#Working with Dicoms\n\n[OsiriX Dicom Viewer](https://www.osirix-viewer.com/)\n\n[Pydicom](https://pydicom.github.io/pydicom/stable/)\n\n[Pydicom documentation](https://pydicom.github.io/pydicom/0.9/)\n\n[Oro Dicom](https://cran.r-project.org/web/packages/oro.dicom/index.html)A package for working with images in R.\n\n[Mango: A viewer for medical research images.](https://mangoviewer.com/)\n\n#Kaggle Inspiration:\n\n[Data Science Bowl 2017](https://www.kaggle.com/competitions/data-science-bowl-2017/overview) In this Competition dataset, you are given over a thousand low-dose CT images from high-risk patients in DICOM format.\n\n[DON'T see like a radiologist! -fastai](https://www.kaggle.com/code/jhoward/don-t-see-like-a-radiologist-fastai) By Jeremy Howard 5y ago\n\n[Basic EDA + Data Visualization](https://www.kaggle.com/code/marcovasquez/basic-eda-data-visualization/notebook) By Marco Vasquez E. - 5y ago\n\n[Standardizing Unusual Dicoms](https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/427217) By Hui Ming Lin\n\n#Deep Learning for MRI\n\nBy Jason A. Polzin\n\nSteps:\n\n\"Image quality inspection: In the first step, the author checked if the given localizer image is suitable to identify the plane for the desired anatomy. This is achieved by using a fiver layer, dyadic reduction regular CNN classification network model (that we call “LocalizerIQ-Net”) to identify slices with relevant anatomy, slices with artifacts and irrelevant slices. If the localizer image is not suitable for ISP, relevant feedback is provided to the scan operator. They used the built-in TensorFlow functions for image manipulation to achieve data augmentation during the training of LocalizerIQ-Net.\"\n\n\"Identification of anatomy coverage: Next, the author located the spatial-extent of the desired anatomy in the localizer images by incorporating a shape-based semantic image segmentation U-Net DL model (called “Coverage-Net”). This helps the next processing steps to be robust to changes in imaging parameter settings across hospitals and clinics as well as changes in shapes and sizes of the patient’s anatomy.\"\n\n\"Identification of precise plane orientation and location: For each desired anatomic structure, the author found the scan-plane that is best suited to image that structure using one or more image segmentation 3D U-Net models (called “Orientation-Net”). Orientation-Net directly segments the desired-planes on localizer images, which is then used to compute the orientation and location.\"\n\n\"The entire process took ~3.5 seconds on a high-performance CPU. The training and testing data for these DL models came from more than 1300 subjects and were acquired using various GE scanner models and field strengths from several clinical sites.\"\n\nhttps://blog.tensorflow.org/2019/03/intelligent-scanning-using-deep-learning.html\n\n#Building Neural Network for Medical Imaging using Deep Learning in Tensorflow \n\nBy Chandan Verma (Kaggler)\n\nSteps below were taken in order to make the data consumable by the CNN:\n\nLoading the Dicom Image, Converting the Pixel values to Hounsfield(HU) units,\nResampling, ROI Generation, Normalization, Zero Centering, Resizing the data\nSplitting the data and label and Creating the Convolutional neural network.\"\n\nConverting Pixel values to Hounsfield units(HU\n\n\"Converting Pixel values to Hounsfield units(HU): This helps us in segregating different parts of the body. Converting to Hounsfield units enable us to extract the lungs which is the area of interest to us in this case. More on Hounsfield units(HU) here. Below is the formula to compute HU from pixel values.\"\n\nhu = pixel_value * slope + intercept \n \nhttps://medium.com/@verma.chandan/building-neural-network-for-medical-imaging-using-deep-learning-in-tensorflow-part-1-ab993b7fb04f", "votes": null }, { "id": "2868787", "postDate": "06/12/2024 16:15:08", "content": "

Thanks! Very useful!

", "rawMarkdown": "Thanks! Very useful!", "votes": null }, { "id": "2888966", "postDate": "06/25/2024 07:06:00", "content": "

Converting pixel values to Hounsfield units (HU) is not applicable for MRI images . Hounsfield units are specific to CT (Computed Tomography) imaging and are used to measure the radiodensity of tissues

", "rawMarkdown": "Converting pixel values to Hounsfield units (HU) is **not applicable** for **MRI images** . Hounsfield units are specific to **CT (Computed Tomography)** imaging and are used to measure the radiodensity of tissues", "votes": null } ], "comments": [ { "id": 2868787, "author_name": "dlt197", "author_url": "", "post_date": "06/12/2024 16:15:08", "content": "

Thanks! Very useful!

", "votes": null, "replies": [] }, { "id": 2888966, "author_name": "vc9408", "author_url": "", "post_date": "06/25/2024 07:06:00", "content": "

Converting pixel values to Hounsfield units (HU) is not applicable for MRI images . Hounsfield units are specific to CT (Computed Tomography) imaging and are used to measure the radiodensity of tissues

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2820940": "#Working with Dicoms\n\n[OsiriX Dicom Viewer](https://www.osirix-viewer.com/)\n\n[Pydicom](https://pydicom.github.io/pydicom/stable/)\n\n[Pydicom documentation](https://pydicom.github.io/pydicom/0.9/)\n\n[Oro Dicom](https://cran.r-project.org/web/packages/oro.dicom/index.html)A package for working with images in R.\n\n[Mango: A viewer for medical research images.](https://mangoviewer.com/)\n\n#Kaggle Inspiration:\n\n[Data Science Bowl 2017](https://www.kaggle.com/competitions/data-science-bowl-2017/overview) In this Competition dataset, you are given over a thousand low-dose CT images from high-risk patients in DICOM format.\n\n[DON'T see like a radiologist! -fastai](https://www.kaggle.com/code/jhoward/don-t-see-like-a-radiologist-fastai) By Jeremy Howard 5y ago\n\n[Basic EDA + Data Visualization](https://www.kaggle.com/code/marcovasquez/basic-eda-data-visualization/notebook) By Marco Vasquez E. - 5y ago\n\n[Standardizing Unusual Dicoms](https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/427217) By Hui Ming Lin\n\n#Deep Learning for MRI\n\nBy Jason A. Polzin\n\nSteps:\n\n\"Image quality inspection: In the first step, the author checked if the given localizer image is suitable to identify the plane for the desired anatomy. This is achieved by using a fiver layer, dyadic reduction regular CNN classification network model (that we call “LocalizerIQ-Net”) to identify slices with relevant anatomy, slices with artifacts and irrelevant slices. If the localizer image is not suitable for ISP, relevant feedback is provided to the scan operator. They used the built-in TensorFlow functions for image manipulation to achieve data augmentation during the training of LocalizerIQ-Net.\"\n\n\"Identification of anatomy coverage: Next, the author located the spatial-extent of the desired anatomy in the localizer images by incorporating a shape-based semantic image segmentation U-Net DL model (called “Coverage-Net”). This helps the next processing steps to be robust to changes in imaging parameter settings across hospitals and clinics as well as changes in shapes and sizes of the patient’s anatomy.\"\n\n\"Identification of precise plane orientation and location: For each desired anatomic structure, the author found the scan-plane that is best suited to image that structure using one or more image segmentation 3D U-Net models (called “Orientation-Net”). Orientation-Net directly segments the desired-planes on localizer images, which is then used to compute the orientation and location.\"\n\n\"The entire process took ~3.5 seconds on a high-performance CPU. The training and testing data for these DL models came from more than 1300 subjects and were acquired using various GE scanner models and field strengths from several clinical sites.\"\n\nhttps://blog.tensorflow.org/2019/03/intelligent-scanning-using-deep-learning.html\n\n#Building Neural Network for Medical Imaging using Deep Learning in Tensorflow \n\nBy Chandan Verma (Kaggler)\n\nSteps below were taken in order to make the data consumable by the CNN:\n\nLoading the Dicom Image, Converting the Pixel values to Hounsfield(HU) units,\nResampling, ROI Generation, Normalization, Zero Centering, Resizing the data\nSplitting the data and label and Creating the Convolutional neural network.\"\n\nConverting Pixel values to Hounsfield units(HU\n\n\"Converting Pixel values to Hounsfield units(HU): This helps us in segregating different parts of the body. Converting to Hounsfield units enable us to extract the lungs which is the area of interest to us in this case. More on Hounsfield units(HU) here. Below is the formula to compute HU from pixel values.\"\n\nhu = pixel_value * slope + intercept \n \nhttps://medium.com/@verma.chandan/building-neural-network-for-medical-imaging-using-deep-learning-in-tensorflow-part-1-ab993b7fb04f", "2868787": "Thanks! Very useful!", "2888966": "Converting pixel values to Hounsfield units (HU) is **not applicable** for **MRI images** . Hounsfield units are specific to **CT (Computed Tomography)** imaging and are used to measure the radiodensity of tissues" }, "source": "meta" }