There is this program mri_synthstrip which inputs a brain MRI file and outputs the same brain MRI file excluding the skull. This is a basic initial step for preprocessing brain MRIs for research.

The program is wonderful and the syntax to skull strip a single MRI is simple if you install the program in your computer:

mri_synthstrip -i input.nii.gz -o stripped.nii.gz

The authors are very nice and they even created a Docker image that you can pull from Dockerhub with docker pull freesurfer/synthstrip

in which case the syntax to skull strip a brain MRI is

docker run -v C:/Users/ch134560/Desktop/MRI_files://data freesurfer/synthstrip -i //data/input.nii.gz -o //data/stripped.nii.gz

What I am trying to do is to create a Docker image that runs through all the MRI files in a directory and skull strips them with synthstrip. I thought this would be easy.

My multistage Dockerfile looks like this, but it does not work because it does not recognize the command in the loop

freesurfer/synthstrip -i "$file" -o "$file"

or

mri_synthstrip -i "$file" -o "$file"

I guess that I do not know how to call a particular command within a unix loop.

#########STAGE 1: CREATE THE MOUNT AND COPY THE MRIs TO BE SKULL-STRIPPED#########

# Use ubuntu as base image
FROM ubuntu AS prepare

# Set working directory
WORKDIR /app

# Copy all MRIs
COPY . .




#########STAGE 2: USE SYNTHSTRIP#########
FROM freesurfer/synthstrip

# Set working directory
WORKDIR /app

# Copy all MRIs
COPY --from=prepare . .

# Create a loop that runs through the MRIs
RUN for file in .; do freesurfer/synthstrip -i "$file" -o "$file"; done

ENTRYPOINT ["ls"]

Other option I thought about is to create a multistage build in which first I import the synthstrip program and then I loop through the MRI files and call synthstrip, but again I do not know exactly how to call the command to run synthstrip from python.

Dockerfile

#########STAGE 1: CREATE THE SYNTHSTRIP CONTAINER###########

# Use synthstrip AS base image
FROM freesurfer/synthstrip as builder

# Set working directory
WORKDIR /app




#########STAGE 2: CREATE THE ANACONDA CONTAINER###########

# Use miniconda as base image
FROM continuumio/miniconda3

# Set working directory
WORKDIR /app

# Copy the initial image
COPY --from=builder . .


# Create the environment
COPY environment.yml .
RUN conda env create -f environment.yml

# Make run commands use the new environment
SHELL ["conda", "run", "-n", "app", "bin/bash", "-c"]

# Code to run when the container is started
COPY main.py .
ENTRYPOINT ["conda", "run", "--no-capture-output", "-n", "app", "python", "main.py"]

main.py file

import os
import subprocess


for MRI in os.listdir('./file_with_MRIs'):
    print(MRI)
    proc = subprocess.Popen(['docker run freesurfer/synthstrip -i ', MRI, ' -o', MRI],
                            stdin = subprocess.PIPE,
                            stdout = subprocess.PIPE,
                            stderr = subprocess.PIPE
                        )

    (out, err) = procmunicate()
    print   (out)

Other option is to create a smaller Docker image that loops through the MRI files and just calls the synthstrip Docker image for each of those and collects the output, but I am not sure how to do that.

Any suggestion or guidance highly appreciated. Thank you very much