Synthesizing Proton-Density Fat Fraction and $R_2^*$ from 2-point Dixon MRI with Generative Machine Learning

TL;DR

This work proposes using a generative machine learning approach to learn PDFF and $R_2^*$ from Dixon MRI and produces synthe-size PDFF and $R_2^*$ maps that show significantly greater correlation with ground-truth than conventional voxel-wise baselines.

Abstract

Magnetic Resonance Imaging (MRI) is the gold standard for measuring fat and iron content non-invasively in the body via measures known as Proton Density Fat Fraction (PDFF) and $R_2^*$, respectively. However, conventional PDFF and $R_2^*$ quantification methods operate on MR images voxel-wise and require at least three measurements to estimate three quantities: water, fat, and $R_2^*$. Alternatively, the two-point Dixon MRI protocol is widely used and fast because it acquires only two measurements; however, these cannot be used to estimate three quantities voxel-wise. Leveraging the fact that neighboring voxels have similar values, we propose using a generative machine learning approach to learn PDFF and $R_2^*$ from Dixon MRI. We use paired Dixon-IDEAL data from UK Biobank in the liver and a Pix2Pix conditional GAN to demonstrate the first large-scale $R_2^*$ imputation from two-point Dixon MRIs. Using our proposed approach, we synthesize PDFF and $R_2^*$ maps that show significantly greater correlation with ground-truth than conventional voxel-wise baselines.

Figures (4)

• Figure 1: Pipeline for training the model to impute PDFF and $R_2^*$. Four channels of Dixon data (in-phase, opposed-phase, water, and fat) were resampled and registered to IDEAL single-slice liver images. Ground-truth PDFF and $R_2^*$ were estimated from IDEAL via nonlinear least squares. A Pix2Pix model was trained to impute PDFF and $R_2^*$ from Dixon.
• Figure 2: Qualitative results on two subjects: one with a) low ($9\%$) liver PDFF, and one with b) high ($32\%$) liver PDFF. The models produce PDFF and $R_2^*$ maps that agree qualitatively with ground-truth (left). The multi-task model has greater error in PDFF estimation in both cases, but the single-task model performs better on the low-PDFF case than the high-PDFF case. The baseline $R_2^*$ estimation method is very inaccurate.
• Figure 3: Mean PDFF and $R_2^*$ for baseline vs. our models and ground-truth (GT). For PDFF (top row), the baseline is less accurate overall. For $R_2^*$, the baseline shows no correlation with ground-truth, while the models show a good correlation. All methods underestimate $R_2^*$ at high values.
• Figure 4: Results on four additional subjects.