using the dimensions of your 2D sprite representing your isometric object doesn't sound like a solid idea to me. a long object along the x axis might has the same sprite width as an object which is wide along the y axis, but both might have different collision boxes.
I would treat those 2 values seperate and calculate collision using bounding boxes that approximate the dimension in isometric space. the offset makes sense because sometimes its not easy to approximate a box for your object. if you have a tree for example, the crown is going to be much bigger than the actual collisionbox which only includes the trunk, but you also want to have the sprite centered within the box and therefor you move the 2D sprite using your offset values.
The image above shows a large sprite (white), with a small collision box (blue) and some offset is used to center the sprite correctly.
Seperating your visual representation from the collision is the way to go in my opinion, even if objects can't be larger then 1 tile of your grid.