Structured errors in reconstruction methods for Non-Cartesian MR data.

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging (MRI)
• Image Reconstruction

Background:

• Traditional analysis of Non-Cartesian MRI reconstruction methods relies on Root Mean Square Error (RMSE).
• RMSE fails to quantify structured errors inherent in the reconstruction process.
• A need exists for more comprehensive error assessment in MRI reconstruction.

Purpose of the Study:

• To introduce and evaluate a novel index for quantifying geometric information loss in MRI reconstruction.
• To compare the performance of Least Squares Non Uniform Fast Fourier Transform (LS-NUFFT) and gridding reconstruction (GR) methods.
• To assess the utility of the geometric information loss index in identifying reconstruction artifacts.

Main Methods:

• Developed a geometric information loss index utilizing the 2D autocorrelation function.
• Compared LS-NUFFT and GR reconstruction methods against Direct Summation (DS) as a reference.
• Calculated both RMSE and geometric information loss on a digital phantom and a hyperpolarized (13)C dataset.

Main Results:

• The geometric information loss index effectively characterized reconstruction errors, even in the presence of noise.
• LS-NUFFT demonstrated lower RMSE compared to GR.
• However, LS-NUFFT reconstructions exhibited more structured errors, indicating significant information loss, observed in both phantom and in vivo data.

Conclusions:

• Evaluating geometric information loss alongside RMSE is crucial for accurate MRI reconstruction performance analysis.
• The geometric information loss index identified that LS-NUFFT discards vital information during reconstruction, a finding masked by its low RMSE.
• This highlights the importance of considering geometric fidelity in MRI reconstruction method assessment.