Artificial neural networks for stiffness estimation in magnetic resonance elastography

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Machine Learning

Background:

• Magnetic Resonance Elastography (MRE) is a non-invasive imaging technique used to measure tissue stiffness.
• Accurate stiffness estimation is crucial for diagnosing and monitoring various diseases, including liver fibrosis and brain conditions.
• Current methods for MRE data inversion can be sensitive to noise and wave interference.

Purpose of the Study:

• To investigate the feasibility of using artificial neural networks (ANNs) for estimating tissue stiffness from MRE data.
• To evaluate the performance of ANNs in MRE data inversion, particularly in terms of accuracy and robustness to noise.

Main Methods:

• ANNs were trained using model-based patterns to estimate stiffness from displacement images obtained via MRE.
• The developed neural network inversions (NNIs) were tested using simulation experiments to assess accuracy under varying conditions of wave interference and noise.
• NNI performance was further evaluated in vivo, comparing results against established direct inversion (DI) methods.

Main Results:

• NNIs demonstrated comparable or superior performance to DI in predicting known stiffness in simulation experiments.
• NNI results showed high correlation with DI results in both liver (R² = 0.974) and brain (R² = 0.915) in vivo.
• NNI identified biologically relevant effects, such as fibrosis stage in the liver and age in the brain, with lower repeatability error in the brain compared to DI.

Conclusions:

• ANNs offer a novel approach for MRE data inversion, providing a powerful tool for stiffness estimation.
• NNIs are highly correlated with DI and effectively detect biologically relevant signals.
• Preliminary findings suggest NNIs may offer improved resistance to noise, warranting further investigation for clinical applications.