NeuroPot (pip package)

NeuroPot is a library which provides automated processing of neuroradiological images, data cleaning and manipulation functionality, machine learning algorithms and transfer learning methods. The preprocessing pipeline works for MPRAGE, but ones following ADNI protocol (MPR-R or MPRAGE REPEAT often DON'T follow the ADNI protocol, please check the orientations before use).

Docker

Build the docker image using:

docker build .
docker tag <image_id> neuropot:version1.0

Create a container:

docker create -it --name neuro1 neuropot:version1.0 bash

Start and attach to a container:

docker start -ai neuro1

This will drop you in the shell!

root@618cac06237d:/usr/src/preprocessor# 
root@618cac06237d:/usr/src/preprocessor# python
Python 3.6.7 (default, Oct 22 2018, 11:32:17) 
[GCC 8.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import neuropot.preprocessing as preproc
>>> 

Quickstart

1. Setup FSL:

Install FSL For MAC OSX, you can download the installer script here

It is important to follow the the shell setup, one way to check whether the shell is not configured is that all fsl applications will appear with a 'fsl-5.0' prefix e.g. 'fsl-5.0-bet' instead of 'bet'.

'commands' package bug. FSL is slightly incompatible with python3. Commands package is depreciated in python3, but imported in a file in FSL (but not used!), so we can remove it:

sudo vim /usr/local/fsl/bin/aff2rigid 

Delete the following line:

from commands import getoutput

3. Setup virtualenv

#Install virtualenv in case not present
pip install virtualenv

#Setup a virtual environment
virtualenv neuro

#Activate the environment
source ./neuro/bin/activate

# You should see the prompt:
# (neuro)$

# To quit the virtualenv
# (neuro)$ deactivate

2. Install the neurpot package via pip:

(neuro)$ pip install neuropot

You should be good to go!

3. Now you can import and use the processing pipeline as follows:

import neuropot.preprocessing as preproc

or

from neuropot.preprocessing import *

You can try the samples given below.

Possible issue with permissions for executing ACPC detect - in case a 'Can't open ACPCAlignment.sh' error, check the permissions for the file and change it to executable:

sudo chmod +x /usr/local/bin/ACPCAlignment.sh
chown <username>:<username> /usr/local/bin/ACPCAlignment.sh

Preprocessing pipeline:

This pipeline requires FSL: Install FSL

• N4 normalization
• ACPC correction
• Skull stripping
• GM Segmentation
• Normalization
• Smoothing

Preprocessing a single file:

import os
import neuropot.preprocessing as preproc

path = os.getcwd()

# N4 normalization
# image_N4_path = preproc.n4_normalization(path+"/input.nii")
# print("Normalization done: ",image_N4_path)

# ACPC correction
image_N4_acpc_path = preproc.acpc_correction(path+"/input.nii")
print("ACPC correction done: ",image_N4_acpc_path)

# Skull stripping
image_N4_acpc_ss_path = preproc.skull_stripping(image_N4_acpc_path)
print("Skull stripping done: ",image_N4_acpc_ss_path)

# GM Segmentation
image_N4_acpc_ss_seg_path = preproc.gm_segmentation(image_N4_acpc_ss_path)
print("GM Segmentation done: ",image_N4_acpc_ss_seg_path)

# Normalization
image_N4_acpc_ss_seg_registered_path = preproc.normalization(image_N4_acpc_ss_seg_path)
print("Normalization done: ",image_N4_acpc_ss_seg_registered_path)

# Smoothing
image_N4_acpc_ss_seg_registered_smooth_path = preproc.smoothing(image_N4_acpc_ss_seg_registered_path)
print("Smoothing done: ",image_N4_acpc_ss_seg_registered_smooth_path)

Preprocessing multiple files:

import os
import neuropot.preprocessing as preproc
from shutil import *

def preprocess(data_dir):
	processed = []
	total = len([file for file in os.listdir(data_dir) if file.endswith(".nii")])

	for idx,file in enumerate(os.listdir(data_dir)):
		if file.endswith(".nii"):
			input_file = os.path.join(data_dir, file)
			print("[%d/%d] Processing %s ... "%(idx+1,total,file),end="", flush=True),
			image_N4_acpc_path = preproc.acpc_correction(input_file)
			image_N4_acpc_ss_path = preproc.skull_stripping(image_N4_acpc_path)
			image_N4_acpc_ss_seg_path = preproc.gm_segmentation(image_N4_acpc_ss_path)
			image_N4_acpc_ss_seg_registered_path = preproc.normalization(image_N4_acpc_ss_seg_path)
			image_N4_acpc_ss_seg_registered_smooth_path = preproc.smoothing(image_N4_acpc_ss_seg_registered_path)
			processed.append(image_N4_acpc_ss_seg_registered_smooth_path)

			copyfile(image_N4_acpc_ss_seg_registered_smooth_path, os.path.join(data_dir,'processed_'+file+'.gz'))
			
			print("[DONE]")

	return processed

def main():
	data_dir = os.path.abspath("./mri")
	processed_files = preprocess(data_dir)
	print(processed_files)

main()