darts 0.0.1

DenseUnet-based Automatic Rapid brain Segmentation

DenseUnet-based Automatic Rapid brain Segmentation (DARTS)

Package

• Our Dense UNet model performs automated segmentation of 102 brain regions within a minute (on a machine with a single GPU). The DARTS package, which utilizes the best-performing model, can be installed using:

pip install DARTS

• Download the pretrained model from here as follows:

gdown https://drive.google.com/uc?id=1ux3WEU9_n7hEy6gjZWI7Y0WL9P47D9kx

Usage

Segmentation requires the use of a GPU and takes approximately one minute. For optimal results, input image size should be 256x256x256 mm³ and in one of the following formats: .mgz, .mgh, .nii.gz or .nii. Follow these steps to perform segmentation:

from DARTS import Segmentation
seg_obj = Segmentation(input_image_path='PATH_TO_INPUT', output_dir='PATH_TO_OUTPUT_DIRECTORY')
seg_out = seg_obj.predict(input_image_path='PATH_TO_INPUT')

Function help:

help(Segmentation)

 |  Methods defined here:
 |  
 |  __init__(self, model_wts_path='./dense_unet_back2front_finetuned.pth')
 |      Initializes the segmentation object (GPU required)
 |      Parameters
 |      ----------
 |      model_wts_path : Path to pre-trained model (default './dense_unet_back2front_finetuned.pth')
 |  
 |  predict(self, input_image_path, output_dir='./')
 |      Segments the input image
 |      Parameters
 |      ----------
 |      input_image_path: Path to input image (can be .mgh, .mgz, .nii or .nii.gz)
 |      output_dir : Path to save segmentation output and volumes (default './')
 |      
 |      Returns
 |      -------
 |      Segmented mask (np.array)
 |  

Project

Here is the paper describing the project and experiments in detail. We pre-trained our model using the Freesurfer segmentations of 1113 subjects available in the Human Connectome Project dataset and fine-tuned the model using 101 manually labeled brain scans from the Mindboggle dataset.

Output segmentation

The output segmentation has 103 labeled segments, which closely resemble the aseg+aparc segmentation protocol of Freesurfer. We also provide region volumes in .csv format. We exclude four brain regions that are not common to a normal brain: White matter and non-white matter hypointentisites, and the left and right frontal and temporal poles. We also exclude left and right bankssts as there is no widely accepted definition of these segments.

Contact

If you have any questions regarding the code, please contact tms598[at]nyu.edu or raise an issue on the github repo.