Tensor denoising of multidimensional MRI data.

Area of Science:

• Medical Imaging
• Signal Processing
• Data Science

Background:

• The Signal-to-Noise Ratio (SNR) is a key limitation in Magnetic Resonance Imaging (MRI).
• Principal Component Analysis (PCA)-based denoising, such as MPPCA, has been used but struggles with small data patches.
• MPPCA's effectiveness relies on a large number of noise singular values relative to signal components, which is often not met in practice.

Purpose of the Study:

• To develop an advanced denoising strategy for high-dimensional data by leveraging its inherent redundancy.
• To introduce a novel tensor-based approach, tensor MPPCA (tMPPCA), to overcome the limitations of matrix-based MPPCA.

Main Methods:

• Introduced tensor MPPCA (tMPPCA), which utilizes the tensor structure of multidimensional data for noise characterization.
• Applied tMPPCA to denoising multidimensional data, including multicontrast acquisitions.
• Compared tMPPCA with traditional matrix-based MPPCA using numerical phantoms and diffusion MRI data.

Main Results:

• tMPPCA demonstrates significantly improved denoising performance compared to matrix-based MPPCA.
• The improvement is particularly notable when dealing with small data patches.
• tMPPCA requires no additional assumptions beyond those of the original MPPCA method.

Conclusions:

• MPPCA denoising technique can be effectively extended to high-dimensional data using a tensor-based approach.
• tMPPCA offers superior performance, especially in scenarios with small data patches, making it beneficial for spatially varying noise.
• This advancement enhances noise reduction capabilities in MRI without introducing artifacts.