Skip to main content
. 2019 Jul 30;13:52. doi: 10.3389/fncom.2019.00052

Figure 1.

Figure 1

Analysis pipeline: images are normalized, then the three types of features were collected. These features are filtered with RFE and PCA, then used to build a random forest model or logistic regression model. Image perturbations are used as an additional filter by including only relatively robust features. The kTSP algorithm used the same feature set to build its predictions.