This is the python version of the "Automated reference tissue normalization of T2-weighted MR images of the prostate using object recognition"

This is an automated method for dual-reference tissue (fat and muscle) normalization of T2-weighted MRI for the prostate.

The method was developed at the CIMORe group at the Norwegian University of Science and Technology (NTNU) in Trondheim, Norway. [https://www.ntnu.edu/isb/cimore]

For detailed information about this method, please read our paper: [https://link.springer.com/article/10.1007%2Fs10334-020-00871-3]

The provided algorithm was developed for research use and was NOT meant to be used in clinic.

pyAutoRef/
├── LICENSE
├── pyproject.toml
├── README.md
├── setup.cfg
├── src/
│   └── pyAutoRef/
│       ├── __init__.py
│       ├── autoref.py
│       ├── pre_processing.py
│       ├── object_detection.py
│       ├── post_processing.py
│       ├── normalization.py
│       ├── utils.py
│       ├── MANIFEST.in
│       └── model.onnx
└── tests/

You can install the package either from pip or using pip or the files in GitHub repository [https://github.com/MohammedSunoqrot/pyAutoRef]

Simply type:

pip install pyAutoRef

• Clone the GitHub repository

  From command line

  git clone https://github.com/MohammedSunoqrot/pyAutoRef.git
  

• Change directory to the clones folder (unzip if needed) and type

  pip install . 
  

This python version is differ than the originally published MATLAB version [https://github.com/ntnu-mr-cancer/AutoRef] in terms:

• It accepts all kind of SimpleITK supported image format.
• It uses YOLOv8 model for object detector instead of ACF.

VERSION 1.0.0, 1.0.1, 1.0.2, 1.0.3, 1.0.4

• YOLOv8 model for object detector trained using the same data and bounding boxes used to train ACF in the original/MATLAB code. No Data deviation, but needed to meantioned.
• The detection was performed on all slices for both fat and muscle.

VERSION 2.0.0 and above

• YOLOv8 model for object detector trained using images from 823 cases (4 centers, 3 centers data is private and 1 is public which is the PROSTATE158 training dataset) in addition to the same data used to train the original detector.
• For detection for both fat and muscle the first 15% and the last 15% of slices were not looked at for detection.
• In detection post-processing, the kernel of opening radius of the detected object was set to 5 pixels.
• In detection post-processing, a morphological erosion step follows the opening has been added with kernel radius of 5 pixels.
• The 95th and 5th percentile of the fat and muscle intensities, respectively, were used to normalize the image.

VERSION 2.1.0 and above

• In case of non-detected fat or muscle objects, the 3 middel slices will be selected, and the normalization will be done using the intensity be calculated using the 95th and 5th percentile of the entire pre-prccessed 3 slices.

In case of using or refering to AutoRef/pyAutoRef, please cite it as:

Sunoqrot, M.R.S., Nketiah, G.A., Selnæs, K.M. et al. Automated reference tissue normalization of T2-weighted MR images of the prostate using object recognition. Magn Reson Mater Phy 34, 309–321 (2021). [https://doi.org/10.1007/s10334-020-00871-3]

To perform AutoRef normalization, you first need to import the autoref function. You can do it by calling from pyAutoRef import autoref

• Parameters:

  • input_image (SimpleITK.Image, str): The input image as SimpleITK.Image OR The file path to the input 3D image (any supported SimpleITK format) or to the DICOM folder.
  • output_image_path (str, optional): The file path to save the normalized output image to any supported SimpleITK format. If None, the image will not be saved.

• Returns:

  • normalized_image (SimpleITK.Image): The normalized 3D image.

• Path to DICOM Series.
• Path to images of all the medical images formats supported by SimpleITK.
• [SimpleITK.Image] (https://simpleitk.org/SimpleITK-Notebooks/01_Image_Basics.html).

DICOM Series is recognized when there is no file extension

Example (input: SiTK Image variable within the code, output: medical image format -Path to write- ):

import SimpleITK as sitk
from pyAutoRef import autoref

file_path = r"C:\Data\Case10_t2.nii.gz"

input_image = sitk.ReadImage(file_path)

output_image_path = r"C:\Data\Case10_t2_normalized.nii.gz"

autoref(input_image, output_image_path)

Example (input: SiTK Image variable within the code, output: SiTK Image variable within the code):

import SimpleITK as sitk
from pyAutoRef import autoref

input_image = sitk.ReadImage(file_path)

normalized_image = autoref(input_image)

Example (input: medical image format, output: SiTK Image variable within the code):

from pyAutoRef import autoref

input_image_path = r"C:\Data\Case10_t2.nii.gz"

normalized_image = autoref(input_image_path)

Example (input: medical image format, output: medical image format):

from pyAutoRef import autoref

input_image_path = r"C:\Data\Case10_t2.nii.gz"
output_image_path = r"C:\Data\Case10_t2_normalized.nii.gz"

autoref(input_image_path, output_image_path)

Example (input: medical image format, output: DICOM Series):

from pyAutoRef import autoref

input_image_path = r"C:\Data\Case10_t2.nii.gz"
output_image_path = r"C:\Data\Case10_t2_normalized"

autoref(input_image_path, output_image_path)

Example (input: DICOM Series, output: medical image format):

from pyAutoRef import autoref

input_image_path = r"C:\Data\Case10_t2"
output_image_path = r"C:\Data\Case10_t2_normalized.nii.gz"

autoref(input_image_path, output_image_path)

Example (input: DICOM Series, output: DICOM Series):

from pyAutoRef import autoref

input_image_path = r"C:\Data\Case10_t2"
output_image_path = r"C:\Data\Case10_t2_normalized"

autoref(input_image_path, output_image_path)