Figure 3 | Scientific Reports

Figure 3

From: Stable and discriminating features are predictive of cancer presence and Gleason grade in radical prostatectomy specimens: a multi-site study

Figure 3

(a) PI-AUC plot for 242 features for 212 studies across four sites for cancer vs. non-cancer classification. Each feature is represented by a dot, color coded according to feature family. On the X-axis is the PI value for each feature in the non-cancerous regions of the four sites. On the Y-axis is the feature’s AUC averaged across 100 iterations of 3-fold cross validation across all patients from all four sites. Shown are 242 features from the global graph (blue), shape (red), disorder (green), sub-graph (yellow), and Haralick (purple) feature families. The optimal high-AUC, low-PI space is dominated by shape features while the low-AUC, high-PI space comprises Haralick and global graph features. (b) ROIs corresponding to features occupying different regions in the PI-AUC space, see Table 2.

Back to article page