I don't think its necessary to use midpoints. Just work out the probability that a<X<b (e.g. 6.5<X<7.5) for each respective class in the normal way (i.e. using the z statistic) then multiply that probability by the total number of trials to work out the expected value.What past paper is that from?