DB-qMRI

Dictionary-based - quantitative magnetic resonance imaging

Description

The dictionary-based learning (DBL) quantitative MRI methods are proposed to bypass inherent magnetic resonance fingerprinting (MRF) limitations: reconstruction time and memory requirement. In particular, we propose a statistical learning to provide both estimates and their confidence levels.

Standard parameter estimation from magnetic resonance fingerprinting (MRF) data is based on matching the MRF signals to their best counterparts in a grid of coupled simulated signals and parameters, referred to as a dictionary Ma et al..

To reach a good accuracy, the matching requires an informative dictionary whose cost, in terms of design, storage and exploration, is rapidly prohibitive for even moderate numbers of parameters. In this package, we propose an implementation of two dictionary-based learning (DBL) approaches made of three steps: 1) a quasi-random sampling strategy to produce efficiently an informative dictionary, 2) a regression model to learn from the dictionary a correspondence between fingerprints and parameters (using either a neural network, e.g. a fully connected network Cohen et al. or an inverse statistical regression model Boux et al.), and 3) the use of this mapping to provide parameter estimates (and their confidence indices for statistical method).

Details about these methods referred to as dictionary-based matching (DBM), dictionary-based deep learning (DB-DL) and dictionary-based statistical learning (DB-SL) can be find in Boux et al..

Configuration

The code has been validated using Matlab R2018 and R2019.

Statistics and Machine Learning Toolbox and Parallel Computing Toolbox toolboxes are required.

Run

Figures from different experiments can be found in the ./figures folder. To generate figures of the paper, the best way is to run the Run.m script:

>> Run

Information about figures are described (see comments) in the Run.m file. Experiments can be launched individually by executing the scripts located in the folder ./Experiments.

The quantification is achieved, running:

>> [Estimation, Parameters] = AnalyzeMRImages(Sequences, Dico, Method)

where Sequences is a 3D or 4D matrix of observed MR signals (the third dimension is the time, others are spatial dimensions), Dico is a structure that represents the dictionary and Method is the strings 'DBM', 'DB-SL' or 'DB-DL' to specify the method to use (see section Description). The fields of Dico are Dico.MRSignals that is a 2D matrix of MR signals (the second dimension is time) and Dico.Parameters.Par is a 2D matrix of parameters (the second dimension is the parameter dimension). Then, note that the first dimensions of Dico.MRSignals and Dico.Parameters.Par must be equals.

Estimation and Parameters are structures. Estimation.Y is the matrix of parameter estimates.

External tools

The ./tools folder contains external toolboxes located in the subfolder having the same name:

• Antoine Deleforge, the GLLiM regression.

• Jakob Asslaender, the NYU_MRF_Recon toolbox reconstructs quantitative maps of arbitrary MRF data with arbitrary k-space trajectories. The tool has been modify in order to take into account any sampling during the dictionary generation.

Contact information

Fabien Boux, fabien.boux@univ-grenoble-alpes.fr