{
  "id": 208505,
  "title": "Better understanding of the problem statement",
  "url": "/competitions/hubmap-kidney-segmentation/discussion/208505",
  "author_name": "Izzy Adesanya",
  "post_date": "2021-01-03T18:46:07.371000",
  "votes": 4,
  "comment_count": 0,
  "views": 0,
  "content": "<p><strong>Commonly used domain terms:</strong></p>\n<p><strong>Capillary</strong>: A capillary is a small blood vessel from 5 to 10 micrometers (μm) in diameter, and having a wall of one endothelial cell thick. They are the smallest blood vessels in the body: they convey blood between the arterioles and venules. These microvessels are the site of the exchange of many substances with the interstitial fluid surrounding them.</p>\n<p><strong>Diffusion Distance</strong>: Intuitively, Diffusion Distance between a pair of points (a, b) can be thought of as following: We create a unit impulse function centered at 'a'(imagine a fixed amount of energy concentrated at a point), and we let it diffuse for a period of time t. We create another impulse function centered at 'b'', and we also let it diffuse for a period time t. In the end, we look at the difference (measured by L2-norm) between the two distributions. And that is our Diffusion Distance.</p>\n<p>Mathematically, this is described by:<br>\n<img src=\"http://www.cs.jhu.edu/~ming/Blog/DiffusionDistance_files/image042.gif\" alt=\"Equation for Diffusion distance\"></p>\n<p><strong>Periodic acid-Schiff (PAS) Stain Microscopy</strong><br>\nPAS is a histology stain that detects complex sugars in tissue sections. Periodic acid is used to break specific bonds within these sugars. The resulting aldehydes react with the Schiff reagent to produce the purple-magenta color exhibited by these images. Glomeruli can be observed as the circular areas of dark stain.</p>\n<p><strong>Glomeruli</strong>: Glomeruli consist of capillaries that facilitate the filtration of waste products out of blood. Normal glomeruli typically range from 100-350 μm in diameter with a roughly spherical shape.</p>\n<p>Glomeruli contain 4 cell types: <br>\na. Parietal epithelial cells that form Bowman’s capsule<br>\nb. Podocytes cover the outer layer of the filtration barrier<br>\nc. Fenestrated endothelial cells that are coated with a glycolipid<br>\nd. Glycoprotein matrix called glycocalyx that are in direct contact with blood and mesangial cells that occupy the space between the capillary blood vessel loops and are stained by the colorimetric histological stain called Periodic acid-Schiff (PAS) stain</p>\n<p><strong>Functional Tissue Unit (FTU)</strong>: a 3D block of cells centered around a capillary, such that each cell in this block is within diffusion distance from any other cell in the same block” (de Bono, 2013) One example of an FTU is the glomerulus found in the outer layer of kidney tissue known as the cortex, which in humans has an area of about 800 mm2 and an average depth of about 9 mm.</p>\n<p><strong>Glomeruli Segmentation Masks</strong><br>\nThe glomeruli segmentation masks are a mix of manually and deep learning (DL) generated annotations in a slightly modified geoJSON format. The JavaScript Object Notation (JSON) file lists all glomeruli identified for each of the 11 + 9 tissue sections. The position and shape of a glomeruli is represented by a set of coordinates. The “detection_score”, present only in the DL generated annotations, is a measure the DL model used during detection.</p>\n<p>Each item in the JSON list is an annotation with the following pertinent fields:</p>\n<ul>\n<li>“geometry”:<ul>\n<li>“geometry/type”: All are “Polygon”</li>\n<li>geometry/coordinates”: A list of each polygon vertex in x,y order</li></ul></li>\n<li>“properties”:<ul>\n<li>“properties/classification”:<br>\n      - “properties/classification/name”: Annotation class (in this case all are “Glomerulus”)</li>\n<li>“properties/measurements”: list of key,value pairs for some quantitative property of the annotation. For annotations generated by the DL model, this includes “detection_score”.</li></ul></li>\n</ul>\n<p><strong>Dataset</strong> (Info is taken from 'About this Competition' section)<br>\nThe dataset is comprised of very large (&gt;500MB - 5GB) TIFF files. The training set has 8, and the public test set has 5. The private test set is larger than the public test set.</p>\n<p>The training set includes annotations in both RLE-encoded and unencoded (JSON) forms. The annotations denote segmentations of glomeruli.</p>\n<p>So basically our task here in this competition is to take PAS kidney image input and identify the segments of glomeruli FTU in the PAS stained microscopy data.</p>\n<p>Please comment on anything that I may have left out or wrongly written. This is a good consolidated discussion for anyone to get started.</p>",
  "messages": [
    {
      "id": 1137198,
      "postDate": "2021-01-03T18:46:07.370Z",
      "content": "<p><strong>Commonly used domain terms:</strong></p>\n<p><strong>Capillary</strong>: A capillary is a small blood vessel from 5 to 10 micrometers (μm) in diameter, and having a wall of one endothelial cell thick. They are the smallest blood vessels in the body: they convey blood between the arterioles and venules. These microvessels are the site of the exchange of many substances with the interstitial fluid surrounding them.</p>\n<p><strong>Diffusion Distance</strong>: Intuitively, Diffusion Distance between a pair of points (a, b) can be thought of as following: We create a unit impulse function centered at 'a'(imagine a fixed amount of energy concentrated at a point), and we let it diffuse for a period of time t. We create another impulse function centered at 'b'', and we also let it diffuse for a period time t. In the end, we look at the difference (measured by L2-norm) between the two distributions. And that is our Diffusion Distance.</p>\n<p>Mathematically, this is described by:<br>\n<img src=\"http://www.cs.jhu.edu/~ming/Blog/DiffusionDistance_files/image042.gif\" alt=\"Equation for Diffusion distance\"></p>\n<p><strong>Periodic acid-Schiff (PAS) Stain Microscopy</strong><br>\nPAS is a histology stain that detects complex sugars in tissue sections. Periodic acid is used to break specific bonds within these sugars. The resulting aldehydes react with the Schiff reagent to produce the purple-magenta color exhibited by these images. Glomeruli can be observed as the circular areas of dark stain.</p>\n<p><strong>Glomeruli</strong>: Glomeruli consist of capillaries that facilitate the filtration of waste products out of blood. Normal glomeruli typically range from 100-350 μm in diameter with a roughly spherical shape.</p>\n<p>Glomeruli contain 4 cell types: <br>\na. Parietal epithelial cells that form Bowman’s capsule<br>\nb. Podocytes cover the outer layer of the filtration barrier<br>\nc. Fenestrated endothelial cells that are coated with a glycolipid<br>\nd. Glycoprotein matrix called glycocalyx that are in direct contact with blood and mesangial cells that occupy the space between the capillary blood vessel loops and are stained by the colorimetric histological stain called Periodic acid-Schiff (PAS) stain</p>\n<p><strong>Functional Tissue Unit (FTU)</strong>: a 3D block of cells centered around a capillary, such that each cell in this block is within diffusion distance from any other cell in the same block” (de Bono, 2013) One example of an FTU is the glomerulus found in the outer layer of kidney tissue known as the cortex, which in humans has an area of about 800 mm2 and an average depth of about 9 mm.</p>\n<p><strong>Glomeruli Segmentation Masks</strong><br>\nThe glomeruli segmentation masks are a mix of manually and deep learning (DL) generated annotations in a slightly modified geoJSON format. The JavaScript Object Notation (JSON) file lists all glomeruli identified for each of the 11 + 9 tissue sections. The position and shape of a glomeruli is represented by a set of coordinates. The “detection_score”, present only in the DL generated annotations, is a measure the DL model used during detection.</p>\n<p>Each item in the JSON list is an annotation with the following pertinent fields:</p>\n<ul>\n<li>“geometry”:<ul>\n<li>“geometry/type”: All are “Polygon”</li>\n<li>geometry/coordinates”: A list of each polygon vertex in x,y order</li></ul></li>\n<li>“properties”:<ul>\n<li>“properties/classification”:<br>\n      - “properties/classification/name”: Annotation class (in this case all are “Glomerulus”)</li>\n<li>“properties/measurements”: list of key,value pairs for some quantitative property of the annotation. For annotations generated by the DL model, this includes “detection_score”.</li></ul></li>\n</ul>\n<p><strong>Dataset</strong> (Info is taken from 'About this Competition' section)<br>\nThe dataset is comprised of very large (&gt;500MB - 5GB) TIFF files. The training set has 8, and the public test set has 5. The private test set is larger than the public test set.</p>\n<p>The training set includes annotations in both RLE-encoded and unencoded (JSON) forms. The annotations denote segmentations of glomeruli.</p>\n<p>So basically our task here in this competition is to take PAS kidney image input and identify the segments of glomeruli FTU in the PAS stained microscopy data.</p>\n<p>Please comment on anything that I may have left out or wrongly written. This is a good consolidated discussion for anyone to get started.</p>",
      "rawMarkdown": "**Commonly used domain terms:**\n\n**Capillary**: A capillary is a small blood vessel from 5 to 10 micrometers (μm) in diameter, and having a wall of one endothelial cell thick. They are the smallest blood vessels in the body: they convey blood between the arterioles and venules. These microvessels are the site of the exchange of many substances with the interstitial fluid surrounding them.\n\n**Diffusion Distance**: Intuitively, Diffusion Distance between a pair of points (a, b) can be thought of as following: We create a unit impulse function centered at 'a'(imagine a fixed amount of energy concentrated at a point), and we let it diffuse for a period of time t. We create another impulse function centered at 'b'', and we also let it diffuse for a period time t. In the end, we look at the difference (measured by L2-norm) between the two distributions. And that is our Diffusion Distance.\n\nMathematically, this is described by:\n![Equation for Diffusion distance](http://www.cs.jhu.edu/~ming/Blog/DiffusionDistance_files/image042.gif)\n\n**Periodic acid-Schiff (PAS) Stain Microscopy**\nPAS is a histology stain that detects complex sugars in tissue sections. Periodic acid is used to break specific bonds within these sugars. The resulting aldehydes react with the Schiff reagent to produce the purple-magenta color exhibited by these images. Glomeruli can be observed as the circular areas of dark stain.\n\n**Glomeruli**: Glomeruli consist of capillaries that facilitate the filtration of waste products out of blood. Normal glomeruli typically range from 100-350 μm in diameter with a roughly spherical shape.\n\nGlomeruli contain 4 cell types: \na. Parietal epithelial cells that form Bowman’s capsule\nb. Podocytes cover the outer layer of the filtration barrier\nc. Fenestrated endothelial cells that are coated with a glycolipid\nd. Glycoprotein matrix called glycocalyx that are in direct contact with blood and mesangial cells that occupy the space between the capillary blood vessel loops and are stained by the colorimetric histological stain called Periodic acid-Schiff (PAS) stain\n \n**Functional Tissue Unit (FTU)**: a 3D block of cells centered around a capillary, such that each cell in this block is within diffusion distance from any other cell in the same block” (de Bono, 2013) One example of an FTU is the glomerulus found in the outer layer of kidney tissue known as the cortex, which in humans has an area of about 800 mm2 and an average depth of about 9 mm.\n\n**Glomeruli Segmentation Masks**\nThe glomeruli segmentation masks are a mix of manually and deep learning (DL) generated annotations in a slightly modified geoJSON format. The JavaScript Object Notation (JSON) file lists all glomeruli identified for each of the 11 + 9 tissue sections. The position and shape of a glomeruli is represented by a set of coordinates. The “detection_score”, present only in the DL generated annotations, is a measure the DL model used during detection.\n\nEach item in the JSON list is an annotation with the following pertinent fields:\n\n-  “geometry”:\n    - “geometry/type”: All are “Polygon”\n    - geometry/coordinates”: A list of each polygon vertex in x,y order\n-  “properties”:\n       - “properties/classification”:\n              - “properties/classification/name”: Annotation class (in this case all are “Glomerulus”)\n       - “properties/measurements”: list of key,value pairs for some quantitative property of the annotation. For annotations generated by the DL model, this includes “detection_score”.\n\n**Dataset** (Info is taken from 'About this Competition' section)\nThe dataset is comprised of very large (>500MB - 5GB) TIFF files. The training set has 8, and the public test set has 5. The private test set is larger than the public test set.\n\nThe training set includes annotations in both RLE-encoded and unencoded (JSON) forms. The annotations denote segmentations of glomeruli.\n\nSo basically our task here in this competition is to take PAS kidney image input and identify the segments of glomeruli FTU in the PAS stained microscopy data.\n\nPlease comment on anything that I may have left out or wrongly written. This is a good consolidated discussion for anyone to get started.",
      "votes": 4
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1137198": "**Commonly used domain terms:**\n\n**Capillary**: A capillary is a small blood vessel from 5 to 10 micrometers (μm) in diameter, and having a wall of one endothelial cell thick. They are the smallest blood vessels in the body: they convey blood between the arterioles and venules. These microvessels are the site of the exchange of many substances with the interstitial fluid surrounding them.\n\n**Diffusion Distance**: Intuitively, Diffusion Distance between a pair of points (a, b) can be thought of as following: We create a unit impulse function centered at 'a'(imagine a fixed amount of energy concentrated at a point), and we let it diffuse for a period of time t. We create another impulse function centered at 'b'', and we also let it diffuse for a period time t. In the end, we look at the difference (measured by L2-norm) between the two distributions. And that is our Diffusion Distance.\n\nMathematically, this is described by:\n![Equation for Diffusion distance](http://www.cs.jhu.edu/~ming/Blog/DiffusionDistance_files/image042.gif)\n\n**Periodic acid-Schiff (PAS) Stain Microscopy**\nPAS is a histology stain that detects complex sugars in tissue sections. Periodic acid is used to break specific bonds within these sugars. The resulting aldehydes react with the Schiff reagent to produce the purple-magenta color exhibited by these images. Glomeruli can be observed as the circular areas of dark stain.\n\n**Glomeruli**: Glomeruli consist of capillaries that facilitate the filtration of waste products out of blood. Normal glomeruli typically range from 100-350 μm in diameter with a roughly spherical shape.\n\nGlomeruli contain 4 cell types: \na. Parietal epithelial cells that form Bowman’s capsule\nb. Podocytes cover the outer layer of the filtration barrier\nc. Fenestrated endothelial cells that are coated with a glycolipid\nd. Glycoprotein matrix called glycocalyx that are in direct contact with blood and mesangial cells that occupy the space between the capillary blood vessel loops and are stained by the colorimetric histological stain called Periodic acid-Schiff (PAS) stain\n \n**Functional Tissue Unit (FTU)**: a 3D block of cells centered around a capillary, such that each cell in this block is within diffusion distance from any other cell in the same block” (de Bono, 2013) One example of an FTU is the glomerulus found in the outer layer of kidney tissue known as the cortex, which in humans has an area of about 800 mm2 and an average depth of about 9 mm.\n\n**Glomeruli Segmentation Masks**\nThe glomeruli segmentation masks are a mix of manually and deep learning (DL) generated annotations in a slightly modified geoJSON format. The JavaScript Object Notation (JSON) file lists all glomeruli identified for each of the 11 + 9 tissue sections. The position and shape of a glomeruli is represented by a set of coordinates. The “detection_score”, present only in the DL generated annotations, is a measure the DL model used during detection.\n\nEach item in the JSON list is an annotation with the following pertinent fields:\n\n-  “geometry”:\n    - “geometry/type”: All are “Polygon”\n    - geometry/coordinates”: A list of each polygon vertex in x,y order\n-  “properties”:\n       - “properties/classification”:\n              - “properties/classification/name”: Annotation class (in this case all are “Glomerulus”)\n       - “properties/measurements”: list of key,value pairs for some quantitative property of the annotation. For annotations generated by the DL model, this includes “detection_score”.\n\n**Dataset** (Info is taken from 'About this Competition' section)\nThe dataset is comprised of very large (>500MB - 5GB) TIFF files. The training set has 8, and the public test set has 5. The private test set is larger than the public test set.\n\nThe training set includes annotations in both RLE-encoded and unencoded (JSON) forms. The annotations denote segmentations of glomeruli.\n\nSo basically our task here in this competition is to take PAS kidney image input and identify the segments of glomeruli FTU in the PAS stained microscopy data.\n\nPlease comment on anything that I may have left out or wrongly written. This is a good consolidated discussion for anyone to get started."
  }
}