A correlation-constrained SNR frequency selection algorithm for magnetic particle imaging reconstruction

System matrix-based reconstruction is a commonly used method in magnetic particle imaging (MPI). However, due to the large memory requirements, lengthy processing times, and ill-posed nature of the system matrix, computational feasibility remains limited to perform reconstruction using the entire matrix. The most common approach involves frequency component selection based on signal-to-noise ratio (SNR), where a limited number of frequency components are chosen for reconstruction. This method accelerates processing while preserving critical reconstruction information. However, selection based solely on SNR risks discarding critical image information, and frequency components with high SNR may exhibit redundant information. Therefore, this paper proposes a correlation-constrained SNR frequency selection algorithm. Its objective is to select high SNR frequency components to ensure the reconstructed image's SNR while prioritizing mutually independent components to better recover image details. Experimental results on real data demonstrate that the proposed algorithm demonstrates superior reconstruction fidelity.