Use of spectral prior in optical tomography has significantly improved accuracy and quality of images, when applied in two-dimensional (2-D) models. However, the size of the problem increases substantially when applied in 3-D. Two methods are presented here that make 3-D spectral imaging computationally feasible. The data-subset approach uses a smaller subset of variable measurement s to reduce the size of the inverse problem. The basic principle consists of using a dynamic criterion to select optimal subset that capture the major changes in the imaging domain. Additionally, the sensitivity matrix is analyzed and made sparse based on a memory requirements (to 8% of full matrix) and provides less tha 2% percent difference in quantification compared to use of full matrices in the image reconstruction.