regardless which type of rendering you use, you always have some sort of viewfrustrum. which can be a box, a frustrum of pyramid (in a usual 3D game), a 2D rectangle in a 2D game, etc. the type of your frustrum is determind by the projectionmatrix.

now you can use your frustrum culling it with the boundingbox of your objects (in 2D for example a 2D rectangle, in 3D a 3D box, or a 3D sphere).

if your frustrum intersects the boundingbox of the object, the object or a part of it is visible - "on screen".

look for some tutorials, how to compute the intersection of a frustrum and a box or whatever.

http://www.realtimerendering.com/int/ here you can find an overview over some intersection computations and where they are listed.