MRI-CUDA

Cuda-accelerated processing of raw MRI data (coming from mridata.org) to generate output images in grayscale.

Data requirements

• Slices must be square (width = height)
• Minimum slice width = 9px
• Maximum slice width = 1024px
• Recommended slice width = 257-1024px
• Output images width is equal to slice width rounded up to the next power of 2.
• Multiple slices per dataset supported
• Multiple channels per slice supported
• Maximum total amount of raw data depends on GPU and CPU memory size (minimum memory required (CPU and GPU) = (8*size*size*num_channels*num_slices) bytes)

How to build (Linux)

Dependencies required

• git
• CUDAToolkit
• CMake (latest version)
• libpugixml
• ISMRMRD lib (installed automatically by the install script)

How to install CMake latest

Install latest version using snap

sudo apt update
sudo apt install snapd
sudo snap install cmake --classic

Installed by default in /snap/bin, needs to manually be included in PATH env.

How to install libpugixml

sudo apt install libpugixml-dev

How to compile the ISMRMRD library and the MRI-CUDA project

Run the script install.sh

In case of errors about other missing libraries, install them manually using apt or other package managers.

How to run

./build/MRI-CUDA <inputFile> <outputDir>

Example run

An example is provided as "ready-to-use". The required data can be downloaded by running the script exampleData.sh (Around 1.6 Gb). The project can then be executed by running the script exampleRun.sh which is already preconfigured.

Developed by

Edoardo Bosi Alfredo Grande

Tested successfully with

CPU:
Intel(R) Xeon(R) CPU @ 2.20GHz (x4)
15 Gb RAM

GPU: 
Nvidia Tesla T4 (16 Gb)