Cell Detection Using Extremal Regions in a Semisupervised Learning Framework

Table 3

Comparison of DICE and detection scores for phase-contrast hela cells. SS refers to our semisupervised approach when the cell candidates are generated using the MSER detector on the intensity images. SS + ILASTIK refers to our semisupervised approach using pixel probability maps from ILASTIK to generate cell candidates using the MSER detector. We evaluate our experiments with and without the contributions of the unlabeled data from the testing dataset.

Method	Detection scores	DICE
Arteta et al. [1]	0.90	0.81
Funke et al. [23]	0.97	0.84
Zhang et al. [21]	0.91	0.80
Our method (SS (only training dataset))	0.94	0.80
Our method (SS (training + testing))	0.95	0.82
Our method (SS + ILASTIK (only training dataset))	0.96	0.86
Our method (SS + ILASTIK (training + testing))	0.97	0.86