Enhanced Parallel MRI Reconstruction Using PCA-Based Machine Learning Denoising
TL;DR
By combining global sparsity constraints (CS) with local low-rank learning (PCA), the method provides a robust and efficient solution for noise-resilient accelerated MRI reconstruction, with strong potential for practical and clinical applications.
Abstract
Compressed Sensing (CS) has been widely adopted in Magnetic Resonance Imaging (MRI) to enable accurate image reconstruction from highly undersampled k-space data, thereby reducing acquisition time. In parallel MRI (pMRI), multi-coil measurements further accelerate imaging but are inherently sensitive to noise amplification and reconstruction artifacts, especially under aggressive undersampling and low Signal-to-Noise Ratio (SNR) conditions. This work presents a machine learning–inspired PCA-based denoising framework integrated into a CS-driven pMRI reconstruction process. Specifically, undersampled k-space data are first reconstructed using a CS-SENSE formulation, after which a patch-based Principal Component Analysis (PCA) is applied to suppress noise by exploiting the low-rank structure of local image patches. Unlike conventional filtering approaches, the proposed PCA denoising adaptively learns dominant signal subspaces, effectively preserving structural details while attenuating both acquisition noise and reconstruction-induced artifacts. Extensive numerical simulations on multi-coil MRI data under various sampling ratios (0.15-0.30) and noise levels (SNR=1-20 dB) demonstrate the effectiveness of the proposed method. The results indicate that the proposed framework effectively enhances reconstruction quality across a wide range of conditions. By combining global sparsity constraints (CS) with local low-rank learning (PCA), the method provides a robust and efficient solution for noise-resilient accelerated MRI reconstruction, with strong potential for practical and clinical applications.