Magnetic Particle Imaging (MPI) is a novel medical imaging modality that can provide excellent sensitivity, contrast and resolution for imaging the spatial distribution of superparamagnetic iron oxide nanoparticles by utilizing their nonlinear magnetization responses. System function reconstruction (SFR) and x-space reconstruction are the two main image reconstruction approaches in MPI. SFR requires time-consuming calibration measurements, which need to be repeated whenever there is a change in scanning parameters or the nanoparticle. In the first part of this thesis, the effects of using mismatched parameters during calibration measurements and imaging in SFR are investigated. Through numerical simulations, MPI signals gathered with different scanning parameters are used for reconstructing images to analyze the effects of parameter changes in image quality in SFR. In contrast to the SFR approach, standard x-space reconstruction does not require calibration measurements. However, the reconstructed images are blurred by the point spread function of the system. In the second part of this thesis, a new learning-based approach is proposed to improve the image quality in x-space reconstructed images. The proposed method learns an end-to-end mapping between the x-space reconstructed blurred images and the underlying nanoparticle distributions. By using numerical simulations, it is shown that the blurring in x-space reconstruction can be significantly reduced with the proposed method.