Adaptive filtering for high resolution magnetic resonance images

Area of Science:

• Medical Imaging
• Signal Processing
• Biomedical Engineering

Background:

• High-resolution magnetic resonance (MR) imaging requires effective noise reduction to improve diagnostic accuracy.
• Traditional adaptive filters often compromise image quality by blurring important edges while enhancing signal-to-noise ratio (SNR).

Purpose of the Study:

• To evaluate novel adaptive filtering techniques for MR images, focusing on improving SNR without sacrificing edge definition.
• To compare the performance of adaptive filtering with noise estimation (AFEN) against established least mean square (LMS) algorithms.

Main Methods:

• Implementation and comparison of five AFEN variations against standard two-dimensional LMS (TDLMS), adaptive filtering with a mean estimator (AFLME), two-dimensional averaged LMS (TDALMS), and two-dimensional median weighted LMS (TDMLMS) algorithms.
• Quantitative and qualitative assessment of SNR improvement and edge preservation in phantom and in vivo inner ear MR images.

Main Results:

• While TDLMS, TDALMS, and TDMLMS filters achieved higher SNR improvement, they resulted in significant edge blurring.
• AFLME and AFEN filters demonstrated approximately a twofold SNR improvement while maintaining superior edge retention.
• AFEN exhibited slightly better performance in both SNR enhancement and edge sharpness compared to AFLME in tested MR images.

Conclusions:

• The proposed AFEN technique offers a promising approach for noise reduction in high-resolution MR imaging.
• AFEN and AFLME provide a better balance between SNR improvement and edge preservation compared to conventional LMS-based filters.
• AFEN shows potential for enhancing the quality of MR images, particularly for applications requiring fine detail, such as inner ear imaging.