Table 2 Diagnostic performances of Models A and B.

From: Inter-vendor performance of deep learning in segmenting acute ischemic lesions on diffusion-weighted imaging: a multicenter study

 

Dice score

Recall

Precision

Model A

Training

0.866 (0.774–0.918)

0.953 (0.851–0.992)

0.736 (0.658–0.780)

Validation

0.835 (0.702–0.899),

0.918 (0.772–0.967)

0.710 (0.596–0.764)

Internal test

0.858 (0.752–0.909)

0.944 (0.828–0.981)

0.639 (0.729–0.773)

External test without TF

0.734 (0.56–0.843)

0.807 (0.616–0.927)

0.624 (0.476–0.716)

External test with TF

0.832 (0.671–0.916)

0.915 (0.738–0.978)

0.707 (0.57–0.779)

Model B

Training

0.896 (0.813–0.940),

0.970 (0.887–0.992)

0.771 (0.7–0.808)

Validation

0.865 (0.745–0.923)

0.943 (0.813–0.983)

0.744 (0.641–0.794)

Internal test

0.857 (0.723–0.921)

0.934 (0.787–0.970)

0.737 (0.622–0.792)

External test without TF

0.756 (0.613–0.851)

0.824 (0.668–0.928)

0.650 (0.527–0.732)

External test with TF

0.846 (0.730–0.902)

0.922 (0.788–0.969)

0.727 (0.638–0.776)

  1. Continuous variables are presented with median and interquartile ranges.
  2. TF transfer learning.
Back to article page