Processing and Analysis of Quantitative Diffusion MRI Data

National Institute Of Biomedical Imaging And Bioengineering, Bethesda

Investigators

Linked publications, trials & patents

Abstract

Diffusion MRI (dMRI) is widely used to investigate structural properties of the brain. In Diffusion Tensor Imaging (DTI) and other dMRI methods, diffusion quantities are computed on a voxel-by-voxel basis from a series of diffusion weighted images (DWIs) acquired with different magnitude and orientation of diffusion sensitization. Therefore, it is of crucial importance to have all DWIs in perfect correspondence, and each image to be artifact-free in order to prevent inaccurate interpretations of analysis results. Investigators in our section have been pioneers in underscoring the importance of the effects of the dMRI processing on the quality of the biological findings that can be achieved. The numerous post-processing strategies we have proposed over the years to improve the quality of dMRI data, have been brought together under the TORTOISE (www.tortoisedti.org, https://github.com/QMICodeBase/TORTOISEV4/) software package framework and released to the public. TORTOISE is a complete dMRI processing & analysis pipeline with different modules tailored for specific tasks. The previous versions encompassed physically-based image registration methods to accomplish the tasks of removing the effects of inter-volume subject motion and eddy current distortion as well as aligning the images to a given template with only one interpolation step, ensuring minimal loss in data quality. TORTOISE has been recently enriched with state-of-art methods to denoise DWI data, remove Gibbs ringing artifacts, freqency drift correction, intra-volume motion correction and outlier detection/replacement. and perform elastic image registration based echo-planar imaging (EPI) distortion correction. TORTOISE can also perform EPI-incuded distortion correction in its DRBUDDI module, which provides morphologically faithful diffusion data. In addition to preprocessing, TORTOISE also provides functionalities for a large range of diffusion tensor estimation (linear and nonlinear regression, robust estimation, free-water compartment estimation, flow estimation). TORTOISE also supports estimation of diffusion models beyond the tensor model and currently provides capabilities to estimate diffusion propagator and derived scalar maps using the Mean Apparent Diffusion MRI (MAPMRI). The latest addition to TORTOISE, i.e. DRTAMAS, enables researchers to perform DTI based population studies by creating population representative DTI atlases and providing strategies to analyze individual deviations from this representative image. TORTOISE has been recently completely revamped with updated programming languages and processing methods to be significantly faster, easier to use while allowing batch programming to process large quantities of data. IT is now fully open-source and can take advantage of GPU architectures for significant a speed-up. Processing speed advantages brought by this version has the potential to enable its direct use in clinical systems through an integration with picture archiving and communication systems (PACS). TORTOISE can be downloaded from https://github.com/QMICodeBase/TORTOISEV4/ . The current user base has exceeded a couple of thousand users and is still growing.