Skip to main content
. Author manuscript; available in PMC: 2019 Nov 2.
Published in final edited form as: IEEE Trans Med Imaging. 2019 Mar 18;38(11):2545–2555. doi: 10.1109/TMI.2019.2905917

TABLE II:

Performance comparison of different approaches using 16×16 patches

The Algorithm Classification Rate on Validation Set Classification Rate on Heldout Set
The selected subset of statistical features [25] 78±1.1% 76.6±1.1%
BoW on raw patches with 20D histograms [25] 80.9±1.1% 79.9±1.1%
The Convlutional-BoW with 20D histograms [26] 81.2±1.1% 79.9±1.1%
The proposed Adversarial-BoW (20D histograms) 84.2±1% 83.8±1%