You have a mixed design 3 x 3 two-way ANOVA. The time interval is in your case just an ordinal variable so it is irrelevant that the intervals are different. In R you can do a standard mixed ANOVA like this:
summary(aov(dv ~ group*surgery + Error(subject/surgery), data))
In case you violate any assumptions you can also conduct a robust mixed design two-way ...
Equal group size is not a requirement for anova (it might be an advantage if you can have it). So, as long as group assignment is still random and representative, as you say it is, there is no reason to do anything.
Just continue to do the analysis.