# PyTorchRadiomics
PyTorch implementation of [PyRadiomics](https://github.com/AIM-Harvard/pyradiomics) Extractor
# Performance Improvement
It can speed up voxel-based features extraction significantly, especially GLCM features.
Using it to extract non-voxel-based features is *NOT* recommended (it is slower).
## Voxel-based Features Extraction Performance Comparison
Intel i9-10900K v.s. RTX 3080 10G (dtype=torch.float64), Size=$16^3$
|Type|CPU Time|Torch Time|Max Abs. Error|Max Rel. Error|
|-|-|-|-|-|
GLCM|636s|23.8s|2.32e-09|7.92e-12|
FirstOrder|4.3s|0.244s|2.84e-14|2.22e-16|
GLRLM|1.71s|0.731s|2.72e-12|8.88e-16|
NGTDM|4.03s|0.398s|3.27e-11|3.99e-15|
# Installation
```
pip install pytorchradiomics
```
# Usage
Only two extra keyword arguments:
1. `device`: `str` or `torch.device`, default: `"cuda"`
2. `dtype`: `torch.dtype`, default: `torch.float64`
Direct usage:
```python
from torchradiomics import (TorchRadiomicsFirstOrder, TorchRadiomicsGLCM,
TorchRadiomicsGLRLM, TorchRadiomicsNGTDM,
inject_torch_radiomics, restore_radiomics)
ext = TorchRadiomicsGLCM(
img_norm, mask_norm,
voxelBased=True, padDistance=kernel,
kernelRadius=kernel, maskedKernel=False, voxelBatch=512,
dtype=torch.float64, # it is default
device="cuda:0",
**get_default_settings())
features = ext.execute()
```
Or use injection to use `RadiomicsFeatureExtractor`:
```python
from radiomics.featureextractor import RadiomicsFeatureExtractor
from torchradiomics import (TorchRadiomicsFirstOrder, TorchRadiomicsGLCM,
TorchRadiomicsGLRLM, TorchRadiomicsNGTDM,
inject_torch_radiomics, restore_radiomics)
inject_torch_radiomics() # replace cpu version with torch version
ext = RadiomicsFeatureExtractor(
voxelBased=True, padDistance=kernel,
kernelRadius=kernel, maskedKernel=False, voxelBatch=512,
dtype=torch.float64, # it is default
device="cuda:0",
**get_default_settings())
ext.execute(img, mask, voxelBased=True)
restore_radiomics() # restore
```
Raw data
{
"_id": null,
"home_page": null,
"name": "pytorchradiomics",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.10",
"maintainer_email": null,
"keywords": "radiomics, cancerimaging, medicalresearch, computationalimaging",
"author": null,
"author_email": "Wingkou <wingkou@outlook.com>",
"download_url": "https://files.pythonhosted.org/packages/a0/2d/462ee778a99a19f9be904a7f3f97cdd90f3ce49ebca9a70c3217f5aa419e/pytorchradiomics-0.0.5.tar.gz",
"platform": null,
"description": "# PyTorchRadiomics\r\n\r\nPyTorch implementation of [PyRadiomics](https://github.com/AIM-Harvard/pyradiomics) Extractor\r\n\r\n# Performance Improvement\r\n\r\nIt can speed up voxel-based features extraction significantly, especially GLCM features.\r\n\r\nUsing it to extract non-voxel-based features is *NOT* recommended (it is slower).\r\n\r\n## Voxel-based Features Extraction Performance Comparison\r\nIntel i9-10900K v.s. RTX 3080 10G (dtype=torch.float64), Size=$16^3$\r\n\r\n|Type|CPU Time|Torch Time|Max Abs. Error|Max Rel. Error|\r\n|-|-|-|-|-|\r\nGLCM|636s|23.8s|2.32e-09|7.92e-12|\r\nFirstOrder|4.3s|0.244s|2.84e-14|2.22e-16|\r\nGLRLM|1.71s|0.731s|2.72e-12|8.88e-16|\r\nNGTDM|4.03s|0.398s|3.27e-11|3.99e-15|\r\n\r\n# Installation\r\n```\r\npip install pytorchradiomics\r\n```\r\n\r\n# Usage\r\n\r\nOnly two extra keyword arguments:\r\n1. `device`: `str` or `torch.device`, default: `\"cuda\"`\r\n2. `dtype`: `torch.dtype`, default: `torch.float64`\r\n\r\nDirect usage:\r\n```python\r\nfrom torchradiomics import (TorchRadiomicsFirstOrder, TorchRadiomicsGLCM,\r\n TorchRadiomicsGLRLM, TorchRadiomicsNGTDM,\r\n inject_torch_radiomics, restore_radiomics)\r\n\r\next = TorchRadiomicsGLCM(\r\n img_norm, mask_norm,\r\n voxelBased=True, padDistance=kernel,\r\n kernelRadius=kernel, maskedKernel=False, voxelBatch=512,\r\n dtype=torch.float64, # it is default\r\n device=\"cuda:0\",\r\n **get_default_settings())\r\n\r\nfeatures = ext.execute()\r\n```\r\n\r\nOr use injection to use `RadiomicsFeatureExtractor`:\r\n\r\n```python\r\nfrom radiomics.featureextractor import RadiomicsFeatureExtractor\r\nfrom torchradiomics import (TorchRadiomicsFirstOrder, TorchRadiomicsGLCM,\r\n TorchRadiomicsGLRLM, TorchRadiomicsNGTDM,\r\n inject_torch_radiomics, restore_radiomics)\r\n\r\ninject_torch_radiomics() # replace cpu version with torch version\r\n\r\next = RadiomicsFeatureExtractor(\r\n voxelBased=True, padDistance=kernel,\r\n kernelRadius=kernel, maskedKernel=False, voxelBatch=512,\r\n dtype=torch.float64, # it is default\r\n device=\"cuda:0\",\r\n **get_default_settings())\r\next.execute(img, mask, voxelBased=True)\r\n\r\nrestore_radiomics() # restore\r\n```\r\n",
"bugtrack_url": null,
"license": null,
"summary": "PyTorch implementation of PyRadiomics Extractor",
"version": "0.0.5",
"project_urls": {
"Github": "https://github.com/lyhyl/pytorchradiomics",
"Homepage": "https://github.com/lyhyl/pytorchradiomics#readme",
"Issues": "https://github.com/lyhyl/pytorchradiomics/issues"
},
"split_keywords": [
"radiomics",
" cancerimaging",
" medicalresearch",
" computationalimaging"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "4c3a09b407489d741333796abd272eda1a1e250917154fe3ccb57cfc360c357c",
"md5": "8ceab26abc9178ee1ee97626def76c1f",
"sha256": "217b890755545c6f64980767d52b6d0d56f32de813b269fff4b5f3c1732f85fe"
},
"downloads": -1,
"filename": "pytorchradiomics-0.0.5-py3-none-any.whl",
"has_sig": false,
"md5_digest": "8ceab26abc9178ee1ee97626def76c1f",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.10",
"size": 29099,
"upload_time": "2024-08-04T16:23:38",
"upload_time_iso_8601": "2024-08-04T16:23:38.174241Z",
"url": "https://files.pythonhosted.org/packages/4c/3a/09b407489d741333796abd272eda1a1e250917154fe3ccb57cfc360c357c/pytorchradiomics-0.0.5-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "a02d462ee778a99a19f9be904a7f3f97cdd90f3ce49ebca9a70c3217f5aa419e",
"md5": "a579f75788dcec0af3855a33cc1f1b2f",
"sha256": "72d75d4faba96d5d9f90b383944e87774275955ef855f6c79da3f201f0dd720e"
},
"downloads": -1,
"filename": "pytorchradiomics-0.0.5.tar.gz",
"has_sig": false,
"md5_digest": "a579f75788dcec0af3855a33cc1f1b2f",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.10",
"size": 22681,
"upload_time": "2024-08-04T16:23:40",
"upload_time_iso_8601": "2024-08-04T16:23:40.312313Z",
"url": "https://files.pythonhosted.org/packages/a0/2d/462ee778a99a19f9be904a7f3f97cdd90f3ce49ebca9a70c3217f5aa419e/pytorchradiomics-0.0.5.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-08-04 16:23:40",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "lyhyl",
"github_project": "pytorchradiomics",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"requirements": [],
"lcname": "pytorchradiomics"
}
JSON
