Table 1 IGTD reduces the local heterogeneity (LH) of image representations compared with REFINED.

From: Converting tabular data into images for deep learning with convolutional neural networks

 

Neighborhood size (p)

LH (IGTD)

LH (REFINED)

Reduction percentage by IGTD

P-value

CCL

3

0.174 (0.023)

0.187 (0.026)

6.38% (7.08%)

2.05E−107

5

0.177 (0.024)

0.187 (0.027)

5.32% (6.90%)

3.37E−87

7

0.179 (0.024)

0.188 (0.027)

4.68% (6.79%)

1.96E−74

9

0.180 (0.024)

0.189 (0.027)

4.33% (6.57%)

4.65E−69

Drug

3

0.051 (0.013)

0.064 (0.017)

19.99% (4.98%)

1.50E−252

5

0.056 (0.014)

0.066 (0.017)

14.64% (4.25%)

2.82E−229

7

0.061 (0.014)

0.069 (0.017)

11.37% (3.92%)

9.87E−199

9

0.067 (0.015)

0.074 (0.018)

8.63% (3.78%)

4.56E−156

  1. In the LH and reduction percentage columns, the number before the parenthesis is the average value obtained across CCLs or drugs, and the number in the parenthesis is the standard deviation. P-value is obtained via two-tail pairwise t-test comparing the LH between IGTD images and REFINED images across CCLs or drugs.
Back to article page