Table 6 20 features which showed an improvement (left column) or worsening (right column) in instability following color normalization.

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

Stabilized by normalization

Destabilized by normalization

Feature

PI change

Feature

PI change

(S) Std. deviation distance ratio

−0.59

(S) Min/max Fourier descriptor 9

0.58

(S) Std. deviation area ratio

−0.59

(S) Median perimeter ratio

0.56

(G) Voronoi chord std. deviation

−0.55

(S) Median Fourier descriptor 8

0.54

(G) Voronoi perimeter std. deviation

−0.54

(S) Median invariant moment 1

0.49

(S) Std. deviation of std. deviation of distance

−0.54

(S) Min/max Fourier descriptor 1

0.46

(G) Voronoi area min/max

−0.46

(D) Mean tensor contrast average

0.45

(G) Voronoi area disorder

−0.45

(H) Mean energy

0.44

(D) Range of tensor information measure 1

−0.43

(S) Median Fourier descriptor 7

0.44

(S) Std. deviation of variance of distance

−0.40

(S) Median invariant moment 2

0.43

(G) Voronoi chord disorder

−0.37

(S) Mean invariant moment 1

0.42

  1. N = 188 non-cancerous regions from 188 patients across all four sites were color normalized to a template and instability across the four sites was calculated before and after normalization. The 10 features with the largest absolute PI change pre- and post-normalization in each direction are shown. A negative PI change signifies a reduction in feature instability. The family of each feature is listed as graph (G), shape (S), disorder (D), sub-graph (SG), and Haralick (H).
Back to article page