Enhancement of the Prefiltered Rotationally Invariant Non-local PCA Algorithm for MRI

TL;DR

This paper enhances the PRI-NL-PCA MRI denoising algorithm by optimizing parameters with particle swarm optimization, combining component filters, and attaching additional filters, showing feasibility and potential for further improvements.

Contribution

It introduces a collaborative approach to PRI-NL-PCA, optimizing parameters and combining filters, and demonstrates its feasibility compared to deep learning methods.

Findings

Original algorithm achieves near-optimal results with parameter tuning.

Adding one NL-PCA filter before and after PRI-NLM improves performance.

The method requires minimal parameter adjustment and has strong generalization potential.

Abstract

Magnetic resonance imaging (MRI) is a non-invasive medical imaging technique offering high-resolution 3D images and valuable insights into human tissue conditions. Even at present, the refinement of denoising methods for MRI remains a crucial concern for improving the quality of the images. This study aims to enhance the prefiltered rotationally invariant non-local principal component analysis (PRI-NL-PCA) algorithm. This paper relaxed the original restriction, using the particle swarm optimization and traversal method to determine the optimal parameters of the algorithm. This paper also combined the component filters of the original algorithm and picked the most suitable combination as the new collaborative algorithm. It was found that the original algorithm has already achieved the best possible outcome, apart from a few threshold parameters that need to be adjusted. The effective way to further enhance the performance is to attach only one NL-PCA filter before and after the pre-filtered rotationally invariant non-local mean (PRI-NLM) filter. Although the performance of the new collaborative algorithm is still a little short of advanced deep learning methods, it shows that the algorithm based on PCA denoising is indeed feasible. It requires only a few parameters to be adjusted, and it is conceivable that they can be determined directly from the image, granting it a strong general capacity for various body parts, and it merits further exploration. An auxiliary tool was also extracted from the new algorithm, encouraging further combination of it with other state-of-the-art methods to further improve their denoising performance.