Skip to main content
. Author manuscript; available in PMC: 2020 Sep 18.
Published in final edited form as: IEEE Access. 2019 May 13;7:64583–64591. doi: 10.1109/access.2019.2916557

Fig. 2.

Fig. 2

Examples of feature maps extracted from the output of 2D CNNs at the end of the first stage: original CT images (top) and their feature maps of CNN trained from scratch (middle) or fine-tuned (bottom). The left, middle, and right panels correspond to the small, middle, and large input sizes, respectively.