Table 2 70 features found in ROIs of the PI-AUC space shown in Fig. 3(b).

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

1: High AUC (>0.94), low PI (<0.1)

2: Low AUC (<0.67), high PI (>0.59)

(S) Mean distance ratio

(G) Voronoi area std. deviation

(S) Mean smoothness

(G) Voronoi chord std. deviation

(S) Mean invariant moment 1

(G) Delaunay side length min/max

(S) Mean invariant moment 2

(G) Delaunay triangle area min/max

(S) Mean invariant moment 4

(G) Delaunay triangle area std. deviation

(S) Std. deviation distance ratio

(G) Voronoi polygon area

(S) Std. deviation smoothness

(SG) Number of end nodes

(S) Std. deviation Fourier descriptor 1

(H) Mean average intensity

(S) Std. deviation Fourier descriptor 2

(H) Mean entropy

(S) Std. deviation Fourier descriptor 3

(H) Std. deviation entropy

(S) Std. deviation Fourier descriptor 4

(H) Mean information measure 1

(S) Std. deviation Fourier descriptor 5

4: High AUC (>0.82), low PI (<0.05) graph

(S) Std. deviation Fourier descriptor 8

(G) Std. deviation neighbors in 40 micron radius

(S) Mean invariant moment 6

(G) Std. deviation neighbors in 60 micron radius

3: Low AUC (<0.5), low PI (<0.1) shape

(G) Std. deviation neighbors in 80 micron radius

(S) Mean fractal dimension

(G) Std. deviation neighbors in 100 micron radius

(S) Mean Fourier descriptor 2

6: High AUC (>0.85) disorder

(S) Mean Fourier descriptor 5

(D) Std. deviation tensor contrast energy

(S) Median invariant moment 7

(D) Mean tensor contrast inverse moment

(S) Median fractal dimension

(D) Std. deviation tensor contrast inverse moment

(S) Median Fourier descriptor 1

(D) Mean tensor contrast average

(S) Median Fourier descriptor 2

(D) Std. deviation tensor contrast average

(S) Min/max invariant moment 6

(D) Std. deviation tensor contrast variance

(S) Min/max invariant moment 7

(D) Mean tensor contrast entropy

(S) Min/max Fourier descriptor 7

(D) Mean tensor contrast entropy

(S) Min/max Fourier descriptor 9

(D) Std. deviation tensor intensity average

5: Mid AUC (<0.64, >0.56) disorder

(D) Mean tensor contrast entropy

(D) Range tensor contrast energy

(D) Std. deviation tensor intensity variance

(D) Range tensor contrast inverse moment

(D) Mean tensor intensity entropy

(D) Range tensor contrast variance

(D) Std. deviation tensor intensity entropy

(D) Range tensor contrast entropy

(D) Mean tensor entropy

(D) Range tensor intensity average

(D) Mean tensor energy

(D) Range tensor intensity variance

(D) Std. deviation tensor energy

(D) Range tensor intensity entropy

(D) Mean tensor correlation

(D) Range tensor entropy

(D) Std. deviation tensor correlation

(D) Range tensor correlation

(D) Mean tensor information measure 2

(D) Range tensor information measure 1

(D) Std. deviation tensor information measure 2

  1. Feature PI and AUC values were found by using all 212 patients from all four sites. PI is calculated across non-cancerous regions and AUC is the mean from 100 iterations of 3-fold cross validation for the cancer vs. non-cancer classification task. Cross validation was performed independently for each site.
Back to article page