INDI: open-source software for processing cardiac diffusion tensor imaging data

Open-source Software Description: BACKGROUND
Cardiac Diffusion Tensor Imaging (cDTI) is a novel imaging technique that provides a unique opportunity to probe the complex and dynamic cardiac muscle architecture in health and disease [1]. There is a clear growing cDTI interest and momentum within the CMR imaging community, with recent clinical studies showing very promising results in cardiomyopathies, congenital heart disease, and myocardial infarction patients [1-6].

cDTI acquires multiple diffusion weighted images (DWIs) that need to be processed to obtain a diffusion tensor and respective metrics for each pixel [7]. Processing in-vivo cardiac diffusion data presents unique challenges, notably respiratory and cardiac motion, and diffusion orientation measures being typically calculated in relation to local cardiac geometry rather than image geometry [8-11]. This typically requires the delineation and identification of heart epi- and endocardial borders and insertion points during the processing workflow, as the orientation of the heart varies from person to person.

In this abstract we present a new open-source, Python-based, cross-platform software tailored to process in-vivo cardiac diffusion tensor data. The software is called INDI (IN-vivo Diffusion Imaging) and is available at (https://github.com/ImperialCollegeLondon/INDI).

WORKFLOW
INDI has a modular design, intended to run primarily from the command line, except for a few modules that require user interaction. It supports spin-echo and STEAM sequences, correcting for spoiler gradients and heart-rate related changes in b-value.
INDI reads cardiac diffusion images (DICOM or NIFTI), and outputs diffusion tensors and respective metrics. The workflow is divided into multiple modules as shown in figure 1.

For a typical one slice dataset (84 DWIs), the analysis time is typically around 7 mins with a standard desktop PC. This includes non-rigid registration (3.7 mins) + other computations (35s) + user interaction. The non-rigid registration requires approx. 6x the processing time of a rigid registration, but it results in improved cDTI parameter maps for scans with considerable inter and intra breath hold motion/Eddy current distortions (figure 2).

A workflow option exists to batch register multiple scans prior to any user input. This can be used for example to pre-register studies in bulk, and subsequently complete the analysis with the necessary manual inputs.
INDI outputs results in a number of common formats, including VTK format that can be used to render results in 3D (figure 3).

DISCUSSION
INDI provides an open-source tool, pushing towards standardisation of cDTI. It will enable large multi-centre studies with non-specialist centres and future translation to clinical adoption.

Modular, open-source well-engineered code written in a language well-established for AI-development provides an ideal foundation for incorporating new AI-modules without rewriting substantial portions of code. Other research groups will be able to contribute and insert their custom modules, allowing for a direct comparison with the established workflow, thereby driving innovation and accelerating progress in the field.

Open-source software access instructions: The source code is hosted on a GitHub repository:
https://github.com/ImperialCollegeLondon/INDI

The default/stable "main" branch should be used to test the software.

The repository README file contains detailed instructions on how to install, and run the software with a test dataset. Please check the run demo video tutorial.