Medical imaging often involves the collective use of information presented in multiple images of an individual, such as images generated through different imaging modalities (X-ray, CT, PET, etc.) The use of a composite of these images may involve image registration to adjust for the variable position and orientation discrepancies of the individual during imaging. However, registration may be complicated by soft tissue deformation between images, and implementations (particularly pure software implementations) of the mathematical models used in image registration may be computationally complex and may require up to several hours. Hardware architectures are presented that apply the mathematical techniques in an accelerated manner, thereby providing near-realtime image registration that may be of particular use for the short timeframe requirements of surgical environments. The composite image generated thereby may be used to target anatomic features of interest during various medical procedures, including surgical procedures. Moreover, such techniques may be applied to computationally difficult image processing techniques, such as the display of a composite image based at least in part on a PET image, which may otherwise be difficult to utilize in a time-sensitive manner such as a surgical setting.