3d change detection using segmentation

Hello

I have two point cloud data (ply files) one for 2016 and anothe for 2018 from a city. I register this point clouds with ICP algorithm nad it returns rmse=4.6 now I segmented this two point clouds according to the euclidean distance now I want to find coordinates that have been changed during 2 years. is there any suggestion for this ?

I think before finde chenged place I should improve accuracy of ICP but I dont have any Idea. can anyone help me?