It does not matter which method you use (the difference is less than 1%), just choose one and use always that. If you primarily work with labelmaps (and use 3D surface for visualization only) then you may prefer labelmap based method; if you mainly work with surfaces then you may choose the volume computed from the generated surface.