We propose for the first time a new strategy to train slice-level classifiers on CT scans based on the descriptors of the adjacent slices along the axis. In particular, each of which is extracted through a convolutional ...