Among other metals, gold nanoparticles (GNPs) play an important role in emerging cancer therapies. The development, safety, and efficacy of these therapies is contingent on the ability to accurately map and quantify GNPs and require a highly sensitive metal-mapping imaging modality that can image GNPs in vivo at relevant concentrations and depths. This dissertation presents an optimized x-ray fluorescence emission tomography (XFET) system and novel joint image reconstruction algorithm to image trace gold for applications in these therapies. XFET is an emerging imaging modality that relies on inducing and detecting x-ray fluorescence to recover the spatial distribution and quantify the density of metal in objects. XFET has numerous advantages over conventional modalities: most notably, its imaging geometry does not require noise-amplifying tomographic image reconstruction, and it does not require the full sinogram associated with conventional tomographic imaging methods. Consequently, XFET may possess the sensitivity and tissue penetration depth required to map metals at concentrations and conditions used in clinical and preclinical studies. In light of these advantages, this dissertation improves and demonstrates XFET capabilities and discusses implications for clinical translation. XFET's imaging mechanism not only allows for direct metal measurement, but also allows for joint estimation of the attenuation map that would otherwise be obtained with a radiation-dose-delivering computed tomography (CT) scan. In the first chapter, we develop a novel joint algorithm to estimate both metal and attenuation maps from emission data alone and show that it outperforms a conventional approach based on linearization. We successfully extend this novel algorithm to the case of an unknown beam attenuation map, demonstrating an accurate joint reconstruction of metal and attenuation maps from emission data alone without prior attenuation knowledge. Image geometry optimization, most notably detector placement, is necessary to obtain accurate metal and attenuation images in XFET. In the second chapter, we use mathematical tools to investigate how detector arrangement affects image quality and joint estimation. We reconstruct metal and attenuation maps using simulated datasets acquired with various detector arrangements. We demonstrate that two parallel detectors provide greatest accuracy but at the cost of reduced isotropic spatial resolution in the attenuation map, informing about optimal detector placement in an imaging task where a full-ring geometry is not feasible. Finally, the third chapter demonstrates proof of benefit by comparing XFET and CT simulations for the task of trace metal mapping. We performed photon-counting CT, energy-integrating CT, and Monte Carlo XFET simulations on two phantoms: the first contained a range of clinically relevant gold concentrations at varied depths in soft tissue, and the second was a realistic numerical mouse phantom. We show that for superficial depths (< 3 cm), XFET outperforms CT for imaging gold concentrations as low as 0.5% by weight. XFET’s detection limit is further improved with additional dose and utilization of XFET's unique partial-field imaging capability. Ultimately, the results of this dissertation improve XFET’s current capabilities and provide the information necessary to predict XFET’s ability to map therapeutic GNPs for informed development, safer treatment, and fewer side effects of cancer therapies.