Difference between revisions of "Cerebellum CNN"

From IACL
Jump to navigation Jump to search
(Update.)
 
(13 intermediate revisions by 2 users not shown)
Line 1: Line 1:
<meta name="title" content="Cerebellum Parcellation with Convolutional Neural Networks"/>
+
<!-- <meta name="title" content="ACAPULCO: Cerebellum Parcellation with Convolutional Neural Networks"/> -->
 
{{h2|Cerebellum Parcellation with Convolutional Neural Networks}}
 
{{h2|Cerebellum Parcellation with Convolutional Neural Networks}}
 
{{TOCright}}
 
{{TOCright}}
This work originally appeared at {{pub|conf=spie2019}} It can be downloaded as a {{iacl|~shuo/data/cerebellum-parcellation.simg|Singularity image}}. The download is 1.3GB. If you use this work please cite:
+
<u>A</u>utomatic <u>c</u>erebellum <u>a</u>natomical <u>p</u>arcellation using <u>U</u>-Net with <u>l</u>ocally <u>c</u>onstrained <u>o</u>ptimization&nbsp;(ACAPULCO) is our current cerebellum parcellation method. The associated publication is:
{{iacl-pub|author=S. Han, Y. He, A. Carass, S.H. Ying, and J.L. Prince|title=Cerebellum parcellation with convolutional neural networks|conf=spie2019|doi=10.1117/12.2512119}}
 
  
{{h3|What the Singularity image can do}}
+
{{iacl-pub| author=S. Han, A. Carass, Y. He, and J.L. Prince| title=Automatic Cerebellum Anatomical Parcellation using U-Net with Locally Constrained Optimization| jrnl=ni| number=218:116819| when=2020| doi=10.1016/j.neuroimage.2020.116819}}
The Singularity performs the following steps:
 
* [https://www.nitrc.org/projects/robex ROBEX] is used to estimate a brain mask. This mask is then smoothed to generate a brain weight image. The mask is not used for skull-stripping. It is only used for the bias field correction below.
 
* N4 from [http://stnava.github.io/ANTs/ ANTs] is used to perform the bias field correction. The bias field is estimated using the weight image calculated above.
 
* The image is rigidly registered to the [http://www.bic.mni.mcgill.ca/ServicesAtlases/ICBM152NLin2009 ICBM2009c] nonlinear symmetric template using the ANTs package.
 
* The cerebellum of an MNI-registered MPRAGE image is parcellated using the method described in "Shuo Han, et al., Cerebellum parcellation with convolutional neural networks, SPIE 2019 Medical Imaging Image Processing".
 
:* Removing the neck should improve the results, such as using "robustfov" from [https://fsl.fmrib.ox.ac.uk/fslcourse/lectures/practicals/intro2/index.html fsl], but it is not done in this Singularity image.
 
* The parcellation is transformed back to the original space using ANTs with the "MultiLabel" interpolation.
 
* Generate .png files of slices in the axial, coronal, and sagittal views for quality control.
 
  
{{h3|Installation}}
 
* Install [https://www.sylabs.io/guides/2.6/user-guide/installation.html Singularity 2.6]
 
  
{{h3|Example Usage}}
+
{|align="center" style="width:75%; border:2px #e99095 solid; background:#ffffff; text-align:left;"
* The Singularity image can only run on '''CPU''' although [https://www.tensorflow.org TensorFlow] and [https://keras.io Keras] are used, because the Singularity image only contains the CPU version of TensorFlow.  
+
|-
 +
| colspan="3" align="center" | '''ACAPULCO'''
 +
|-
 +
<!--
 +
| {{iacl|~shuo/data/acapulco_022.tar.gz|Docker image CPU}}
 +
| 1.8GB
 +
|-
 +
| {{iacl|~shuo/data/acapulco_022_gpu.tar.gz|Docker image GPU}}
 +
| 2.5GB
 +
|-
 +
-->
 +
| {{iacl|~shuo/data/acapulco_030.sif|Singularity image CPU}}
 +
| Feb 27  2022
 +
| 1.8GB
 +
|-
 +
| {{iacl|~shuo/data/acapulco_030_gpu.sif|Singularity image GPU}}
 +
| Feb 27  2022
 +
| 2.5GB
 +
|-
 +
|}
  
* Assume that the Singularity image is ~/cerebellum-parcellation.simg, the image to parcellate is image.nii.gz, and the output folder is ~/output
 
singularity run ~/cerebellum-parcellation.simg -i ~/image.nii.gz -o ~/output
 
  
* If the image.nii.gz is under /path/to/image which is not under your home directory
+
If you use it please cite the above paper. If you have any questions, please email Shuo Han at <code>shan50 _AT_ jhu.edu</code>.
  
singularity run -B /path/to/image:/mnt ~/cerebellum-parcellation.simg -i /mnt/image.nii.gz -o /mnt/output
 
  
* Print help
+
{{h3|Instructions}}
singularity run ~/cerebellum-parcellation.simg -h
 
  
{{h3|File structure}}
+
The instructions are available [https://gitlab.com/shuohan/keras-unet-cerebellum/-/blob/master/README.md here].
  
The processing will create subfolders '''"n4"''', '''"mni"''', '''"parc"''', and '''"pics"'' under the output folder. The final parcellation result is directly under the output folder.
+
{{h3| Delineation protocols: region names and voxel values}}
* '''"*n4.nii.gz"''' is the bias field corrected image
 
* '''"*n4_mni.nii.gz"''' is the image in the MNI space
 
* '''"*n4_mni_seg.nii.gz"''' is the parcellation from the deep networks
 
* '''"*n4_mni_seg_post.nii.gz"''' is the post-processed result.
 
* '''"*n4_mni_seg_post_inverse.nii.gz"''' is the final post-processed result in the original space.
 
* The subfolder '''"pics/final"''' contains the axial, coronal, and sagittal slices of the file '''"*n4_mni_seg_post_inverse.nii.gz"'''.
 
  
{{h3|Cerebellum labels}}
+
* Adult training data:
    "12":  "Corpus_Medullare",
+
"Background":             0,
    "33": "Left_Lobules_I-III",
+
  "Corpus Medullare":      12,  
    "36": "Right_Lobules_I-III",
+
  "Left I-III":            33,  
    "43":  "Left_Lobule_IV",
+
  "Right I-III":            36,  
    "46":  "Right_Lobule_IV",
+
"Left IV":               43,
    "53":  "Left_Lobule_V",
+
  "Rigt IV":                46,  
    "56":  "Right_Lobule_V",
+
"Left V":                 53,
    "63":  "Left_Lobule_VI",
+
  "Right V":                56,  
    "60":  "Vermis_VI",
+
"Vermis VI":             60,
    "66":  "Right_Lobule_VI",
+
  "Left VI":                63,  
    "73":  "Left_Lobule_VIIAf",
+
"Right VI":               66,
    "76":  "Right_Lobule_VIIAf",
+
  "Vermis VII":            70,  
    "74":  "Left_Lobule_VIIAt",
+
"Left Crus I":           73,
    "70":  "Vermis_VII",
+
  "Left Crus II":          74,  
    "77":  "Right_Lobule_VIIAt",
+
"Left VIIB":             75,
    "75":  "Left_Lobule_VIIB",
+
  "Right Crus I":          76,  
    "78":  "Right_Lobule_VIIB",
+
"Right Crus II":         77,
    "83":  "Left_Lobule_VIIIA",
+
  "Right VIIB":            78,  
    "80":  "Vermis_VIII",
+
"Vermis VIII":           80,
    "86":  "Right_Lobule_VIIIA",
+
  "Left VIIIA":            83,  
    "84":  "Left_Lobule_VIIIB",
+
"Left VIIIB":             84,
    "87":  "Right_Lobule_VIIIB",
+
  "Right VIIIA":            86,  
    "93": "Left_Lobule_IX",
+
"Right VIIIB":           87,
    "90": "Vermis_IX",
+
  "Vermis IX":              90,  
    "96": "Right_Lobule_IX",
+
"Left IX":               93,
    "103": "Left_Lobule_X",
+
  "Right IX":              96,  
    "100": "Vermis_X",
+
"Vermis X":             100,
    "106": "Right_Lobule_X"
+
  "Left X":                103,
 +
"Right X":               106
 +
 
 +
 
 +
 
 +
* Pediatric training data:
 +
  "Background":              0,
 +
"Right Lobules I-V":       3,
 +
  "Right Lobule VI":        4,
 +
"Right Crus I":           5,
 +
  "Right Crus II/VIIB":      6,
 +
"Right Lobule VIII":       7,
 +
  "Right Lobule IX":        8,
 +
"Right Lobule X":         9,
 +
  "Left Lobules I-V":      10,
 +
"Left Lobule VI":         11,
 +
  "Left Crus I":            12,
 +
"Left Crus II/VIIB":     13,
 +
  "Left Lobule VIII":      14,
 +
"Left Lobule IX":         15,
 +
"Left Lobule X":         16,
 +
"Vermis I-V":             17,
 +
"Vermis VI-VII":         18,
 +
"Vermis VIII-X":         19,
 +
"Corpus Medullare":       20

Latest revision as of 18:19, 2 November 2022

Cerebellum Parcellation with Convolutional Neural Networks

Automatic cerebellum anatomical parcellation using U-Net with locally constrained optimization (ACAPULCO) is our current cerebellum parcellation method. The associated publication is:

  • S. Han, A. Carass, Y. He, and J.L. Prince, "Automatic Cerebellum Anatomical Parcellation using U-Net with Locally Constrained Optimization", NeuroImage, 218:116819, 2020. (doi)


ACAPULCO
Singularity image CPU Feb 27 2022 1.8GB
Singularity image GPU Feb 27 2022 2.5GB


If you use it please cite the above paper. If you have any questions, please email Shuo Han at shan50 _AT_ jhu.edu.


Instructions

The instructions are available here.

Delineation protocols: region names and voxel values

  • Adult training data:
"Background":             0,
"Corpus Medullare":       12, 
"Left I-III":             33, 
"Right I-III":            36, 
"Left IV":                43, 
"Rigt IV":                46, 
"Left V":                 53, 
"Right V":                56, 
"Vermis VI":              60, 
"Left VI":                63, 
"Right VI":               66, 
"Vermis VII":             70, 
"Left Crus I":            73, 
"Left Crus II":           74, 
"Left VIIB":              75, 
"Right Crus I":           76, 
"Right Crus II":          77, 
"Right VIIB":             78, 
"Vermis VIII":            80, 
"Left VIIIA":             83, 
"Left VIIIB":             84, 
"Right VIIIA":            86, 
"Right VIIIB":            87, 
"Vermis IX":              90, 
"Left IX":                93, 
"Right IX":               96, 
"Vermis X":              100,
"Left X":                103,
"Right X":               106


  • Pediatric training data:
"Background":              0,
"Right Lobules I-V":       3,
"Right Lobule VI":         4,
"Right Crus I":            5,
"Right Crus II/VIIB":      6,
"Right Lobule VIII":       7,
"Right Lobule IX":         8,
"Right Lobule X":          9,
"Left Lobules I-V":       10,
"Left Lobule VI":         11,
"Left Crus I":            12,
"Left Crus II/VIIB":      13,
"Left Lobule VIII":       14,
"Left Lobule IX":         15,
"Left Lobule X":          16,
"Vermis I-V":             17,
"Vermis VI-VII":          18,
"Vermis VIII-X":          19,
"Corpus Medullare":       20