Deep Learning-Based Water-Fat Separation from Dual-Echo Chemical Shift-Encoded Imaging

Abstract

Conventional water–fat separation approaches suffer long computational times and are prone to water/fat swaps. To solve these problems, we propose a deep learning-based dual-echo water–fat separation method. With IRB approval, raw data from 68 pediatric clinically indicated dual echo scans were analyzed, corresponding to 19382 contrast-enhanced images. A densely connected hierarchical convolutional network was constructed, in which dual-echo images and corresponding echo times were used as input and water/fat images obtained using the projected power method were regarded as references. Models were trained and tested using knee images with 8-fold cross validation and validated on out-of-distribution data from the ankle, foot, and arm. Using the proposed method, the average computational time for a volumetric dataset with ~400 slices was reduced from 10 min to under one minute. High fidelity was achieved (correlation coefficient of 0.9969, l1 error of 0.0381, SSIM of 0.9740, pSNR of 58.6876) and water/fat swaps were mitigated. I is of particular interest that metal artifacts were substantially reduced, even when the training set contained no images with metallic implants. Using the models trained with only contrast-enhanced images, water/fat images were predicted from non-contrast-enhanced images with high fidelity. The proposed water–fat separation method has been demonstrated to be fast, robust, and has the added capability to compensate for metal artifacts.

QSM Throughout the Body

“…Recent advances in deep learning (DL) MR image reconstruction brought forth novel methods for water-fat separation and to estimate chemical-shift-free field maps. [79][80][81][82][83][84][85] DL reconstruction is fast, eliminates the need for initial guess and reduces dependence on acquisition parameters while improving signal-to-noise ratio (SNR) of reconstructed maps. 79,80 However, this approach requires large training sets to sufficiently capture data characteristics and prevent generalization errors.…”

AI in MRI: Computational Frameworks for a Faster, Optimized, and Automated Imaging Workflow

“…The ability to derive meaningful tissue-characterizing images from raw data is yet another appealing application of AI in MRI. In this issue, Wu et al designed a convolutional neural network (CNN) for the synthesis of water/fat images from dual- (instead of multi-) echo images [ 52 ]. In addition to the high fidelity shown in the output images, the proposed method demonstrated a 10-fold acceleration in computation time and a generalization ability to unseen organs and metal-artifact-containing images.…”

De-Aliasing and Accelerated Sparse Magnetic Resonance Image Reconstruction Using Fully Dense CNN with Attention Gates

“…∑ j∈subbands I C →N,j ; T →N,j |s N,j (10) where R →N,j , and T →N,j represent the sub-bands of the reconstructed and target images, respectively; S N,j defines a realization for a specific image; and C →N,j expresses N elements of random field C j that specifies the coefficient of the sub-band, j. The evaluation result of VIFP is indicated as values between 0 and 1, similar to the SSIM.…”

“…Deep learning (DL) has been successfully adopted in medical image processing [9][10][11][12] and deals with the obstacles of iterative CS reconstruction [13,14]. Efficient reconstruction can be performed without repeatedly processing after properly training a DL model.…”