Disparity between estimated & actual points towards either bad statistics on the data, or good statistics, but the data is skewed which is causing bad parameter sniffing for some values.

For posting the execution plan, screen captures don't really work well. You can export the execution plan as a *.sqlplan file and post that here. That way all available information is there. Also, I'd post both plans, the good one & the bad one so that comparisons can be made.