An uncertainty aided framework for learning based liverT_1ρmapping and analysis.

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Medical Imaging Analysis
• Artificial Intelligence in Healthcare

Background:

• Quantitative T1ρ imaging shows promise for assessing liver pathologies.
• Deep learning accelerates quantitative T1ρ imaging, but clinical application requires uncertainty estimation.
• Uncertainty quantification is crucial for reliable AI-based quantitative imaging in clinical settings.

Purpose of the Study:

• To develop a probabilistic deep learning framework for uncertainty-aware quantitative T1ρ mapping in the liver.
• To improve the accuracy and reliability of AI-driven T1ρ quantification for liver disease assessment.
• To utilize uncertainty maps for refining T1ρ mapping and identifying unreliable data points.

Main Methods:

• Proposed a parametric map refinement approach using probabilistic deep learning for T1ρ mapping.
• Trained the model to simultaneously estimate T1ρ values and their associated uncertainty.
• Utilized uncertainty maps to spatially weight the training of an improved T1ρ mapping network.

Main Results:

• The learning-based map refinement achieved a relative mapping error below 3% with simultaneous uncertainty estimation.
• The estimated uncertainty accurately reflected the actual error levels in T1ρ quantification.
• Using uncertainty maps reduced the relative T1ρ mapping error to 2.60% and effectively removed unreliable pixels.

Conclusions:

• The proposed approach offers a trustworthy, learning-based quantitative MRI system for liver T1ρ mapping.
• Uncertainty estimation enhances the reliability of AI-driven quantitative imaging in clinical practice.
• This method has significant potential for improving the diagnosis and management of liver pathologies.