Technologies are described herein for generating a diagnostic three dimensional image for a patient. Some example technologies may obtain a sequence of multiple images of the patient using an imaging modality device. The technologies may estimate a registration vector for each image based on a motion function and an image transformation function. Each image may be defined by a measurement noise added to the image transformation function operating on the registration vector with respect to a reference image. The registration vector may be a function of a breathing motion of a prior registration vector added to a transition noise value. The technologies may estimate motion parameters based on the registration vector. The technologies may iteratively refine the registration vector and the motion parameters. The technologies may generate the diagnostic three dimensional image of the patient using the registration vector for each image and the motion parameters.
