Also for what it's worth, the group I am working with a my University does research in graphics and visualization, and some of their grad students are specialists in medical visualization.
I'm currently working for that group developing a software as a service platform for their volume rendering software, and from which to continually add new algorithms and features they develop, so that their collaborators at medical schools can have easy access.
The collaborators are radiologists and biomedical engineers. They have a lab where they invent and build new machines for imaging. They generate complex data sets and rely on computer scientists to design algorithms and build software which allows you view and interact with the data in real time.
Computer-assisted diagnosis (CAD) - the development of algorithms for systematic analysis of images was very hot at the RSNA meeting a few years ago - and I suspect that it will be for years to come. This field will incorporate a lot of image registration work, as researchers and doctors will be needing to bring together information from multiple imaging modalities and use that for a more personalized, or patient-specific approach to medicine.
GPU-based Monte Carlo simulations have also been hot over the last few years. We use these a lot in radiation oncology.
Modelling of cancer progression and response to treatment is also a big field, although the computational side of things isn't really the bottleneck right now (it's more the availability of quality biological data).
I was recently offered an internship involving the design and programming of heart pumps - the programming part is a great practical example of the CS and medical fields working together. I was already in the middle of a project, though, and couldn't accept this enticing offer :(
If you are interested in more direct uses of computation, there are roles in both diagnostics and therapeutics.
I don't know much about therapeutics, but as Choppy said, on the diagnostics side there is CAD, including image segmentation, registration, and the recognition of lesions. CAD is widely used in mammography but not so widely used elsewhere yet. Digital pathology, which is one of the most complicated fields to digitize, is still in its infancy.
In CT and nuclear medicine, there is innovation ongoing in the areas of reconstruction algorithms.