Qualitative and quantitative evaluation of six algorithms for correcting intensity nonuniformity effects

Area of Science:

• Medical Imaging
• Neuroimaging
• Image Processing

Background:

• Intensity nonuniformity is a common artifact in magnetic resonance imaging (MRI).
• Numerous nonuniformity-correction (NUC) algorithms exist, but their comparative performance is not well-established.
• Selecting the appropriate NUC algorithm is crucial for accurate neuroscientific applications.

Purpose of the Study:

• To evaluate and compare the performance of six NUC algorithms.
• To assess accuracy, precision, and stability of NUC methods using simulated and real MRI data.
• To investigate the influence of intersubject variability and scanner differences (1.5T and 3.0T) on algorithm performance.

Main Methods:

• Utilized simulated and real MRI data, including repeat scans from the same subject.
• Assessed NUC algorithms using phantom studies to measure correlation between extracted and applied nonuniformity.
• Evaluated algorithm stability through iterative application to corrected outputs.
• Compared performance across different subjects and MRI scanner field strengths.

Main Results:

• Phantom studies showed highest correlation of corrected nonuniformity in the transaxial (left-to-right) direction and lowest in the axial (top-to-bottom) direction.
• Two out of six algorithms demonstrated high stability under iterative correction.
• Locally adaptive NUC methods generally showed superior performance compared to nonadaptive methods.
• Performance differences were observed related to intersubject variability and scanner field strength.

Conclusions:

• No single NUC algorithm is ideal for all situations.
• Locally adaptive methods offer a promising approach for MRI nonuniformity correction.
• Algorithm selection should consider the specific neuroscientific application and potential sources of variability.