The purpose of this study was to develop and validate a technique for unsealed source radiotherapy planning that combines the segmentation and registration tasks of single-photon emission tomography (SPECT) and computed tomography (CT) datasets. The segmentation task is automated by an atlas registration approach that takes advantage of a hybrid scheme using a diffeomorphic demons algorithm to warp a standard template to the patient's CT. To overcome the lack of common anatomical features between the CT and SPECT datasets, registration is achieved through a narrow band approach that matches liver contours in the CT with the gradients of the SPECT dataset. Deposited dose is then computed from the SPECT dataset using a convolution operation with tracer-specific deposition kernels. Automatic segmentation showed good agreement with manual contouring, measured using the dice similarity coefficient and ranging from 0.72 to 0.87 for the liver, 0.47 to 0.93 for the kidneys, and 0.74 to 0.83 for the spinal cord. The narrow band registration achieved variations of less 0.5 mm translation and 1° rotation, as measured with convergence analysis. With the proposed combined segmentation-registration technique, the uncertainty of soft-tissue target localization is greatly reduced, ensuring accurate therapy planning.