The imaging capability of magnetic resonance force microscopy has been demonstrated with a 25 nm resolution spin-noise image of a single electron spin and a 4-to-10 nm resolution spin-noise image of few hundred nuclear spins. A recent demonstration of hyperpolarization in a mechanically detected magnetic resonance experiment opens up new possibilities for utilizing polarized spin signal for nanoscale imaging. In this thesis, we present our development of new image reconstruction methods for both nuclear spin ensembles and individual electron spins, and our signal-to-noise analysis to determine the quickest path to a highest-resolution image. First, we speed up the image-reconstruction process in spin noise imaging experiments by revising the iterative Landweber algorithm, and employing the Fourier convolution theorem to reduce the reconstruction time from three days to a few hours. We invent a fast reconstruction scheme to further speed up the process, combining Fourier deconvolution and Tikhonov regularization. This reconstruction scheme yields a three-dimensional image within a few minutes of calculation, but compromises the image signal-to-noise ratio by creating false peaks in the resulting image. This problem of false peaks is resolved by a third reconstruction scheme based on Bayesian Monte Carlo calculation, capable of producing error bars for each pixel in the image. Second, we develop a new detection and image-reconstruction protocol for individual electron spins attached to a single biomolecule. The image-reconstruction algorithm combines a fast Fourier deconvolution with a rigorous Markov-chain reverse Monte Carlo calculation. We demonstrate this new protocol via numerical simulations for a protein molecule with two nitroxide spin labels attached. The sparsely located electron spins allow three-dimensional coordinates of the spin labels to be reconstructed from a two-dimensional scanned signal map, reducing the required signal-acquisition time by 64-to-128 times compared to the three-dimensional scanning in previous imaging experiments. Finally, we derive the signal-to-noise ratios for Fourier-encoded imaging experiments in two polarization limits: detecting the spin fluctuations in the spin-noise limit, and detecting the Curie-law signal from hyperpolarized nuclear spins. Our results indicate that hyperpolarization, when done right, can help reduce the signal-acquisition time in nanoscale imaging experiments and bring the three-dimensional image resolution to the sub-nanometer scale.