MR fingerprinting Deep RecOnstruction NEtwork (DRONE)

TL;DR

A novel fast method for reconstruction of multi‐dimensional MR fingerprinting (MRF) data using deep learning methods and it is shown that this method can be used to solve the challenge of integrating 3D image recognition and 3D handwriting analysis.

Abstract

PURPOSE: Demonstrate a novel fast method for reconstruction of multi-dimensional MR Fingerprinting (MRF) data using Deep Learning methods. METHODS: A neural network (NN) is defined using the TensorFlow framework and trained on simulated MRF data computed using the Bloch equations. The accuracy of the NN reconstruction of noisy data is compared to conventional MRF template matching as a function of training data size, and quantified in a both simulated numerical brain phantom data and acquired data from the ISMRM/NIST phantom. The utility of the method is demonstrated in a healthy subject in vivo at 1.5 T. RESULTS: Network training required 10 minutes and once trained, data reconstruction required approximately 10 ms. Reconstruction of simulated brain data using the NN resulted in a root-mean-square error (RMSE) of 3.5 ms for T1 and 7.8 ms for T2. The RMSE for the NN trained on sparse dictionaries was approximately 6 fold lower for T1 and 2 fold lower for T2 than conventional MRF dot-product dictionary matching on the same dictionaries. Phantom measurements yielded good agreement (R2=0.99) between the T1 and T2 estimated by the NN and reference values from the ISMRM/NIST phantom. CONCLUSION: Reconstruction of MRF data with a NN is accurate, 300 fold faster and more robust to noise and undersampling than conventional MRF dictionary matching.

Figures (7)

• Figure 1: Schematic of the reconstruction approach used in this study. MRI data acquired with the optimized MRF EPI sequence is fed voxelwise to a four layer neural network containing two$300 \times 300$ hidden layers. The network is trained by a dictionary generated using Bloch equation simulations with the tanh and sigmoid functions used as activation functions of the first and last hidden layers respectively. The network then outputs the underlying
• Figure 2: Evolution of the training cost as a function of the epoch number shown on a log scale for improved visualization. The 1000 epochs used in the training required approximately 10 minutes on an Nvidia K80 GPU though the majority of the cost reduction occurred within the initial 400 epochs implying that training time may be reduced without adversely impacting the results.
• Figure 3: Shown is a comparison between the training (true)$\mathrm{T}_{1}$ and $\mathrm{T}_{2}$ and those reconstructed by
• Figure 4: True and reconstructed$\mathrm{T}_{1} / \mathrm{T}_{2}$ images from the numerical brain phantom shown on a common ms scale and the associated error map. Note the close agreement between the reconstructed and true maps. The $\mathrm{T}_{1}$ and $\mathrm{T}_{2}$ RMSEs of 3.5 and 7.8 ms respectively are shown inset in white in the error map.
• Figure 5: Mean (circle) and standard deviation (whiskers) of the reconstruction RMSE of noisy data using conventional dictionary matching (red) or the proposed NN (blue) for undersampled dictionaries. The full dictionary was undersampled by the undersampling factors shown and was used to either directly match the noisy data or to train a NN which was then used to reconstruct the noisy data. Note the lower error for the NN reconstruction despite increased undersampling.