No abstract

No abstract

No abstract

INTRODUCTION:Compressed sensing has been shown to be a useful method for reducing the acquisition time of a system matrix in Magnetic Particle Imaging (MPI) [1][2][3]. The system matrix is acquired by discretising the field of view (FOV) and performing a measurement on each of the resulting spatial points. The combination of the spectra of those measurements is the system matrix. Using methods of compressed sensing, the FOV can be undersampled by acquiring a subset of the spatial points and reconstructing the system matrix in a sparse domain. In the last years, the system matrix of a 2D MPI scanning device by Philips research has been exploited [4]. In this work, proven methods have been applied on a new dataset that has been acquired using a 2D single sided MPI scanning device [5][6][7].METHODS: A 2D system matrix has been acquired with the scanner presented by Gräfe et al. [7] with the exciting frequencies of 25.252 kHz and 26.042 kHz in x-and y-direction, respectively. The FOV of 30 x 30 mm² has been discretised into 30 x 30 pixels. To simulate the undersampling, 80 % of the acquired spatial points have been discarded randomly. Figure 1 shows the undersampling pattern with black spots marking the discarded measurements. The spectra of the remaining 180 measurements have been used to form the undersampled system matrix. The rows of the undersampled system matrix are transformed into sparse domain using the DCT-I. In sparse domain, they are reconstructed by the FISTA [8]. The regularisation parameter = 5e-4 has been determined to be the one that introduces the lowest normalized rootmean-square error (NRMSE). Before back-transforming the reconstructed data, the 95 % smallest coefficients in sparse domain are discarded for each row of the system matrix. RESULTS: Even with= 5e-4 being the parameter that introduces the lowest error, the NRMSE is about twice as high as with the dataset presented in [4]. By undersampling the system matrix with a factor of 0.20, the reconstructed system functions get noisy (see Figure 2). Both the background and the signal itself are disrupted. After discarding the 95 % lowest values in sparse domain, a noise reduction is visible. CONCLUSION:Even though it is possible to undersample the system matrix of the single sided MPI scanning device, its reconstruction introduces errors. As proposed in [9], the system matrix can be denoised after the reconstruction by discarding small values in sparse domain that represent noise and are located mainly in high frequency information for low frequency components [1]. Although the reconstruction result improves, it is still worse than with the data in [2]. The magnetic field of a single-sided scanning device is highly inhomogeneous, see [6]. The lower the distance to the scanner, the higher is the signal. Therefore, weighting the undersampling pattern in favor of spatial points with small distance to the scanner may improve the reconstruction results of the undersampled system matrix. ACKNOWLEDGMENTS:The BMBF (Grant Number 13GW0069A) supports this...

No abstract