Sparse MR Image Reconstruction Considering Rician Noise Models: A CNN Approach.

Area of Science:

• Medical Imaging
• Signal Processing
• Machine Learning

Background:

• Compressive sensing (CS) in magnetic resonance (MR) imaging faces challenges with high computational time for image reconstruction and Rician noise removal.
• Existing methods often model MR image noise as Gaussian, limiting the effectiveness of advanced noise models like Rician within the CS paradigm.

Purpose of the Study:

• To develop a novel framework for high-speed, high-quality MR image reconstruction from sparse k-space data corrupted by Rician noise.
• To address the limitations of current CS techniques in terms of computational time and noise modeling.

Main Methods:

• A convolutional neural network (CNN) is employed for denoising MR images affected by Rician noise.
• The algorithm processes similar image patches in groups to extract local features, leveraging signal similarities.
• CNN training on a GPU using the Convolutional Architecture for Fast Feature Embedding framework significantly reduces runtime for online reconstruction.

Main Results:

• The proposed CNN-based framework achieves high-speed reconstruction with excellent visual quality, even at 20-fold undersampling.
• The method demonstrates high accuracy and consistent peak signal-to-noise ratio.
• It eliminates the need for noise level optimization and prediction, a key advantage over existing techniques.

Conclusions:

• The developed framework offers a significant improvement for MR image reconstruction, particularly in scenarios with sparse k-space data and Rician noise.
• The high-speed reconstruction capability makes it suitable for online applications, including wireless data transmission and remote health monitoring.