Skip to main content
. Author manuscript; available in PMC: 2018 Jun 1.
Published in final edited form as: Inf Process Med Imaging. 2017 May 23;10265:170–183. doi: 10.1007/978-3-319-59050-9_14

Fig. 3.

Fig. 3

Performance of our method. (a) compares KL to BOW while (b) illustrates the importance of the divergence choice. While KL outperforms BOW, the HE divergence has a negative R2, indicating that LLE is not able to accurately approximate the structure of the latent space. (c) The ROC curve indicates the ability of our subject level interpretation method to detect abnormal patches in comparison with random selection. The box in the right shows the zoomed in area of the curve for FPR < 10%.