FIGURE 1.

Examples of targeting. Each row of images is from a single patient. (A) MRI T2^∗ fast-acquisition gradient-echo input images. (B) Ground truth images of supervised semantic segmentation training or validation. (C) Semantic segmentation results and automatic targeting results from a deep learning network trained by 62 non-augmented images. (D) Automatic targeting results superimposed on the original image. The automatic targets shown in these figures are for subthalamic nucleus deep brain stimulation. Deep learning-based semantic segmentation and targeting adaptability is shown for various anatomical variations, including right and left asymmetries in the second and third rows, the large inter-red nuclei distance in the fourth row, and large red nuclei in the fifth row.