Having helped pathology folks before (with absolutely no pathology knowledge), I would say testing and refactoring. Most of the people involved on these projects are not CS folk, so they write code that "works". Tons of repetitive, poorly tested, and usually inefficient approaches. Obviously you can't help much with domain specifics but a clean code base goes a long way. Or just fixing algorithmic or non-optimal approaches. For instance, fixing roundabout python code to use list comprehension, which is more idiomatic and more performant than most other raw python approaches.