Automated quantification of T1 and T2 relaxation times in liver mpMRI using deep learning: a sequence-adaptive approach

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Quantitative MRI

Background:

• Liver multiparametric MRI (mpMRI) provides valuable diagnostic information.
• Quantitative analysis of T1 and T2 relaxation times is crucial for liver phenotyping.
• Current methods for liver mpMRI analysis can be time-consuming and operator-dependent.

Purpose of the Study:

• To evaluate a deep learning-based, sequence-adaptive algorithm for automated liver mpMRI assessment.
• To validate the algorithm's performance in quantifying total and segmental T1 and T2 relaxation times across different populations.
• To assess the reliability and potential clinical utility of automated liver mpMRI analysis.

Main Methods:

• A neural network was trained to segment liver parenchyma and vessels from noncontrast T1-weighted images.
• The algorithm was co-registered to T1 and T2 relaxation time maps (noncontrast and contrast-enhanced) for analysis.
• External validation was performed on independent cohorts of patients with liver fibrosis and healthy controls, comparing automated to manual measurements.

Main Results:

• Deep learning-based automated segmentation showed excellent agreement with manual measurements for both segmental (ICC=0.95) and total liver analysis (ICC=0.97).
• Median differences between automated and manual measurements were minimal: 1.8 ms for noncontrast T1, 2.0 ms for contrast-enhanced T1, and 0.3 ms for T2.
• The algorithm demonstrated high reliability across diverse patient populations and liver segments.

Conclusions:

• Automated quantification of liver mpMRI using deep learning is highly effective and reliable.
• The sequence-adaptive design allows for comprehensive quantitative analysis without sequence-specific training.
• This automated approach has the potential to enhance clinical decision-making and individual liver phenotyping.