Adaptive and robust frequency selection framework in calibration-based magnetic particle imaging reconstruction.

Abstract

Frequency selection is a crucial step for calibration-based magnetic particle imaging (MPI) reconstruction, enabling improvement in computational efficiency and noise suppression. Current methods combine signal-to-noise ratio (SNR) feature with a selection threshold. However, the selection threshold determination is experience-dependent, and the utilization of the system matrix (SM) and the imaging phantom signal is insufficient. To suppress these issues, an adaptive and robust frequency selection framework (AR-FSF) is proposed, including three modules: (i) Velocity-corrected feature calculation, which limits feature calculation to the calibration points with high field-free-region velocity, (ii) Adaptive threshold calculation, which adaptively calculates the noise level using the feature spectrum, (iii) Forward-backward selection, which selects high-SNR frequency components for both SM and imaging phantom for reconstruction. Signal experiments validate the effectiveness and the robustness of the introduced modules respectively. Reconstruction experiments further validate that the AR-FSF method can provide a simple and robust frequency selection process for reconstruction. In experiments using in-house data, the AR-FSF method provides suitable frequency components for fast and high-quality imaging, requiring a minimum reconstruction time of 4.5% compare to current methods. The proposed AR-FSF method effectively simplifies the frequency selection process, enabling adaptive selection of frequency component for different phantoms, thereby achieving fast and high-quality reconstruction. The AR-FSF method simplifies the frequency component selection process and can be widely applied in calibration-based MPI reconstruction, laying a methodological foundation for future biomedical applications.