Figure 1.

(A) Schematic representation of the localizer architecture and its training. The CT slices in input are axial slices taken from the same CT scan. Z¹ and Z² are the axial coordinates of each input slice, respectively. The red square symbolizes the binary cross entropy loss function used during training. (B) A use-case of the localizer. Each axial slice is processed through the network to generate a score. A linear relation between the scores and the axial coordinate is estimated. Cropping of the thorax and abdomen is done based on the anatomical scores, and corresponding axial slice.