Improved pixel-by-pixel MRI R2* relaxometry by nonlocal means

Abstract

Purpose: To investigate the feasibility of improving MRI R2* mapping by filtering the images before curve-fitting.

Methods: Pixel-by-pixel curve-fitting for the quantification of MRI relaxometry remains a challenge for low signal-to-noise ratio images. By computing the weighted mean of spatially adjacent pixels, the low-pass Gaussian (LPG) filter can suppress the noise but at the expense of blurring. By assigning high weights to pixels with similar neighborhood patches, the nonlocal means (NLM) algorithm can reduce noise while retaining intrinsic signals, however, its potential has not been explored in pixel-by-pixel MRI relaxometry, and in this study, we aimed to investigate the impact of the LPG and the NLM filtering on decay signals and MRI R2* mapping. These two filtering methods were compared on both simulated and in vivo data.

Results: Both LPG and NLM algorithms produces R2* maps with decreased root-mean-square-errors. The LPG filter blurs edges of R2* maps while the NLM algorithm preserves details well. The NLM consistently yields R2* mapping with smaller errors than the LPG filtering in all cases.

Conclusion: Pixel-by-pixel fitting can skew MRI relaxometry. The NLM outperforms the conventional LPG filter and has the potential to provide more accurate pixel-by-pixel MRI relaxometry for improved tissue characterization.

Keywords: MRI relaxometry; R2* mapping; noise; nonlocal means; pixel-by-pixel.