First, to avoid aggregation as you had we use a medium containing methylcellulose (1%). With the methylcellulose the media will be thicker and the cells/spheres will not move much.
Second, we plate several different quantity of cells to start the assay (500, 1000, 5000 and 20000). With 500 we will find more variability and with 20000 more aggregation. So, we now are plating always 5000 cells.
Third, we use a plate reader to take 69 pictures per well in triplicate (yes, that is possible). We use the imageXpress software to take the pictures and them to analyse it. Of course that it is not easy at the beginning since you will need to set up and teach the software to analyse. Nonetheless, after this hard work the process will be relatively quick and automatised. Relatively because it takes around 16 min per plate to get the pictures and around 19 min per plate to analyse.
Hope that this answer will help you and did not come to late...