In theory you could create ROIs in each timepoint separately (rather than the whole timeseries together) and as long as the cells don’t leave the frame and there are the same number, you will have the same number of ROIs. But, obviously they will re-arrange and the ROIs will be numbered differently (I presume it is always indexed left-right, top-bottom).
It would be best to use a tracking plugin such as TrackMate. With this plugin you can automatically track the same cell as it moves, it will produce spots (circle ROIs) for each cell at each timepoint and link the spots for a single cell by a track. However, the spots are usually just circles and so you lose the segmented ROI that you need to do your intensity analysis. Although, you can export the spots from TrackMate and I could envisage using the information (which spots are in what track, and where the spots are located in x/y coordinates) to index your spots and segment them and link your data together like that.
I did consider using the ROI x/y coordinates you already have to do a nearest neighbour analysis (found this plugin). I presume with this plugin (not tested) that you can find the two nearest ROIs from adjacent timepoints and use that information to get your mean intensity measure over time from the specified ROIs. However, presumably it will make some incorrect links between cells that are touching or move past each other, this is why TrackMate would be better.
EDIT: TrackMate also has some additional extras, like extracting the intensity of each spot automatically. But, I’m not sure if there is an easy way to segment the cells before doing intensity analysis.