The Diffuse Optical Tomography (DOT) has received considerable attention in the recent years in the field of biomedical imaging and disease detection. However, imaging through highly diffusive medium is a challenge and stability is always an issue due to the inverse problem. Here a non-linear continous wave (CW) semi-analytic reconstruction method is discussed that used curved-beam paths for tomographic imaging with no assumption on inclusion, unlike iterative methods.The non-linear Rosenbrock's function is used to approximate the paths followed by majority photons as curved ones. The modified Beer-Lambert Law (MBLL) in proposed differntial form is used to calculate the absorption coefficient of all the avilable photon paths. The computed values are back-projected along these channels and serve as the basis for image reconstruction without solving the inverse problem. For three-dimensional (3-D) imaging, measurements at three different depths covering the entire depth of the phantom are taken. These slices are stacked together followed by interpolation to form the volumetric image of the phantom to complete the tomographic imaging in its true sense. Numerical simulations, wax phantom experiments with different geometries and contrast are carried out with satisfactory results. This semi-analytic reconstruction method is simple and efficient and is suitable for real-time applications that reqiure fast absorption image reconstruction. The method is compared with the Greedy algorithms for further validation. Also, different performance evaluation matrices are estimated to assess the accuracy of the reconstructed images and the results are rather satisfactory.