Pixelwise Uncertainty Quantification of Accelerated MRI Reconstruction.

Area of Science:

• Medical Imaging
• Machine Learning
• Image Reconstruction

Background:

• Parallel imaging accelerates MRI scans but increases reconstruction artifacts as acceleration factor rises.
• Current clinical practice uses conservative acceleration factors due to the lack of automated methods to assess reconstruction quality.

Purpose of the Study:

• To develop a general framework for pixel-wise uncertainty quantification in parallel MRI.
• To enable automatic identification of unreliable regions in undersampled reconstructions without ground-truth data.

Main Methods:

• Integration of conformal quantile regression with image reconstruction techniques.
• Development of a Variational Network for end-to-end reconstruction of Cartesian undersampled brain and knee data (acceleration factors 2-10).

Main Results:

• High agreement (Pearson correlation >90% at 4x acceleration) between predicted uncertainty maps and true reconstruction error.
• Uncertainty maps accurately capture the magnitude and spatial distribution of reconstruction errors, highlighting artifacts and pathologies.
• Demonstrated superior performance compared to heuristic residual-based uncertainty estimation.

Conclusions:

• The proposed framework provides statistically rigorous, pixel-wise uncertainty quantification for parallel MRI.
• Enables reliable assessment of reconstruction quality without fully-sampled reference images.
• Paves the way for adaptive MRI protocols balancing scan time and diagnostic accuracy.