The computational problem for solving Cryo-EM structures is an interesting one. The current state-of-the-art approach is essentially an EM algorithm for classifying and estimating the orientations of the images, but it utilises a fairly brute-force approach (and thus quite computationally expensive). There is probably a lot of room here for further improvements, such as better or faster approximations, or using different statistical/machine learning approaches, as well as incorporating more chemical/structural knowledge into the process.

The software I'm most familiar with is relion, which is open source: https://github.com/3dem/relion