Table 1 The set of 6 gland lumen features selected by the elastic-net Cox regression model as being most prognostic of BCR on the n = 214 patient training set.

From: Computer extracted gland features from H&E predicts prostate cancer recurrence comparably to a genomic companion diagnostic test: a large multi-site study

Feature name

Hazard ratio

Shape: Mean Fourier descriptor 3

1.002

Shape: Standard deviation of the first invariant moment

0.932

Shape: Median mean/max imum ratio of radius

1.100

Shape: 5%/95% Fourier descriptor 6

0.968

Shape: 5%/95% Fourier descriptor 9

0.929

Sub-Graph: Kurtosis of edge length

0.977

  1. The hazard ratio of feature x is \({e}^{{\beta }_{x}}\) where β is the vector of weights from the fitted Cox regression model. The hazard ratios are shown here reflect the risk of an increase of one standard deviation in feature value on the training set. A hazard ratio less than 1 implies that an increase in that feature’s value is associated with a reduced risk of BCR, while a hazard ratio greater than 1 implies the opposite
Back to article page