Interpretable Dimension Reduction for MRI Channel Suppression.

Area of Science:

• Medical Imaging
• Machine Learning
• Signal Processing

Background:

• Channel suppression in MRI reduces redundant data from receiver coils, accelerating image reconstruction.
• Principal Component Analysis (PCA) is used for channel suppression but lacks interpretability due to full channel contribution.
• Interpretability in machine learning is increasingly vital in radiology.

Purpose of the Study:

• To enhance the interpretability of PCA-based channel suppression in MRI.
• To develop a sparse PCA method for channel suppression that sets most coil loadings to zero.

Main Methods:

• Channel suppression formulated as a nonlinear eigenvalue problem.
• Utilized the inverse power method for solving the eigenvalue problem.
• Proposed a sparse PCA approach to achieve zero loadings for most coils.

Main Results:

• Sparse PCA-based channel suppression demonstrated improved interpretability through sparse channels.
• Achieved enhanced reconstruction quality compared to standard PCA.
• Maintained similar reconstruction times as standard PCA.

Conclusions:

• Sparse PCA offers a more interpretable alternative for MRI channel suppression.
• The proposed method improves both interpretability and reconstruction quality in MRI.
• This technique addresses the need for interpretable machine learning in medical imaging.