CrossAlignNet: a self-supervised feature learning framework for 3D point cloud understanding | Synapse