The animations below illustrate our iterative instance segmentation approach to vertebra segmentation. A red box indicates the current field of view of the network. Voxels that are detected within the field of view are shown in bright red/pink. Once a vertebra is fully segmented, it is shown in a different color and the network moves on to segment the next vertebra. Incompletely visible vertebrae are also shown in these animations, but are normally discarded.

Low-dose chest CT

This network was trained to follow the spine upwards, from feet to head.

Spine CT

This network was trained to follow the spine downwards, from head to feet.