Automated liver magnetic resonance elastography quality control and liver stiffness measurement using deep learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Hepatology

Background:

• Magnetic resonance elastography (MRE) is crucial for liver fibrosis staging.
• Current MRE methods face challenges in quality control (QC) and measurement variability.
• Automating QC and liver stiffness measurement (LSM) is needed for clinical utility.

Purpose of the Study:

• To develop and validate a fully automated deep learning (DL) method for liver MRE QC.
• To automate liver stiffness measurement (LSM) using DL.
• To assess the performance and efficiency of the DL approach compared to manual methods.

Main Methods:

• A retrospective study utilized 897 MRE slices from 69 patients.
• A SqueezeNet-based model was trained for artifact detection in MRE images.
• A U-Net segmentation model was employed for liver segmentation and LSM on diagnostic-quality slices.

Main Results:

• The DL QC model achieved high accuracy (0.958), precision (0.982), and recall (0.886).
• The DL-assisted LSM demonstrated a low mean error of 1.9% ± 4.6% compared to the reference standard.
• Automated LSM was completed in under 1 second per slice, a significant improvement over manual methods (20 minutes).

Conclusions:

• The automated DL-based approach for liver MRE quality classification, segmentation, and LSM shows high performance.
• This method has the potential for clinical adoption to streamline liver fibrosis assessment.
• The DL approach offers a promising solution for overcoming QC challenges and measurement variability in MRE.