Deep reinforcement learning using a multi-scale agent with a normalized reward strategy for automatic cephalometric landmark detection | IEEE Conference Publication | IEEE Xplore