Table 1 Evaluation of the E-U-Net performance.

From: Live-dead assay on unlabeled cells using phase imaging with computational specificity

 

Ground Truth

Live (n = 1973)

Dead (n = 246)

PICS

Live

98.8%

2.4%

Dead

1.2%

97.6%

Evaluation

Precision

99.6%

91.2%

Recall

98.8%

97.6%

F1 Score

99.1%

94.3%

  1. An object-based accuracy metric is used to estimate the deep learning prediction by comparing the dominant semantic label of HeLa cell nuclei with the ground truth. The entries of the confusion are normalized with respect to number of cells in each class.
Back to article page