Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
January 30, 2019Journal of Magnetic Resonance Imaging

Deep learning-based water-fat separation correlated highly with conventional model-based methods (R2 ≥ 0.97, P<0.001) and provided a 14% higher signal-to-noise ratio (P<0.001).

View Full Paper
Ask AI
Bookmark
Share

Why the study?

Cardiac MRI water-fat separation has become complicated by multiparametric analytic models, optimization methods, and the lack of a unified framework for diverse source data.

Does deep learning via a convolutional neural network improve water-fat separation and parameter mapping in cardiac MRI compared to conventional model-based methods?

Comparison

U-Net deep learning water-fat separation vs graph cut reference standard

Design

Retrospective study

Key result

Deep learning-based water-fat separation correlated highly with conventional model-based methods (R2 ≥ 0.97, P<0.001) and provided a 14% higher signal-to-noise ratio (P<0.001).

Authors

JGJames W. GoldfarbJCJason CraftJCJie Cao

Discussion

Loading...

Member takes

Overview

Feasibility of multiparametric water-fat separation shown in limited data; leaves open unified frameworks for diverse MR sources.

Study Design

Type

Observational (n=90)

Structured PICO

Does deep learning via a convolutional neural network improve water-fat separation and parameter mapping in cardiac MRI compared to conventional model-based methods?

P
Population
90 cardiac MR examinations from normal controls and subjects with acute, subacute, and chronic myocardial infarction.
E
Exposure
U-Net deep learning (DL) for water-fat separation and parametric mapping with single and multiecho, complex, and magnitude inputs
C
Comparator
Conventional model-based water-fat separation (graph cut method with R2*, off-resonance correction, and a multipeak fat spectrum)
O
Outcome
Image structural similarity, quantitative proton density fat fraction (PDFF), R2*, and off-resonance quantitative valuessurrogate

Main Result

Effect estimate: R2 ≥ 0.97

p-value: p=<0.001

Deep learning-based water-fat separation in cardiac MRI is feasible and provides quantitative results comparable to conventional methods with a higher signal-to-noise ratio.

Cite This Study

Goldfarb et al. (2019) conducted an observational in Myocardial infarction (n=90). Deep learning (U-Net) water-fat separation vs. Conventional model-based water-fat separation (graph cut method) was evaluated on Correlation of predicted values for R2*, off-resonance, water, and fat signal intensities (R2 ≥ 0.97, p=<0.001). Deep learning-based water-fat separation correlated highly with conventional model-based methods (R2 ≥ 0.97, P<0.001) and provided a 14% higher signal-to-noise ratio (P<0.001).

synapsesocial.com/papers/6a63775a84b804377828ba57https://doi.org/10.1002/jmri.26658
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1Fast water/fat T2 and PDFF mapping via multiple overlapping‑echo detachment acquisition and deep learning reconstruction2025 · 1 citations
  2. 2Non-iterative and uncertainty-aware MRI-based liver fat estimation using an unsupervised deep learning method2025 · 1 citations
  3. 3In Vivo Evaluation of a Novel Deep Learning-based MR Image Reconstruction for Liver Fat Quantification2024
  4. 4Robust Water-Fat Separation in Nasal MRI Using Hybrid Model and Data Driven Network2025
  5. 5Fat-water separation in low SNR MRI using Graph Cut and mixed magnitude-complex re-fitting with Tukey's Bi-weighting2025