As a new non-invasive medical imaging technology, diffuse optical tomography (DOT) has received considerable attention that can provide vast quantities of functional information of tissues. The reconstruction problem of DOT is highly ill-posed, meaning that a small error in the measurement data can bring about drastic errors of the reconstruction optical properties. In this paper, the shape-based image reconstruction algorithm of DOT is proposed for reducing the ill-poseness under the assumption that the optical properties of target region distribute uniformly. Since some human organs and tumors can be simplified as an ellipsoid, in this paper, the shape of the inhomogeneity is described as an ellipsoid. In the forward problem, the boundary element method (BEM) is implemented to solve the continuous wave diffusion equation (DE). By the use of the ellipsoid parametric method, the description of the shape, location and optical properties of the inhomogeneity, and the value of the background could be realized with only a small number of parameters. In the inverse calculation, a Levenberg-Marquardt algorithm with line searching is implemented to solve the underlying nonlinear least-squares problem. Simulation results show that the algorithm developed in this paper is effective in reducing the ill-poseness and robust to the noise.