Cell Detection Using Extremal Regions in a Semisupervised Learning Framework

Table 1

Comparison of inference methods for phase-contrast hela cells. SS + ILASTIK refers to our semisupervised method using pixel probability maps from ILASTIK. The unlabeled data has contributions from both the training and testing images.

	SS + ILASTIK (greedy inference)			SS + ILASTIK (bottom-up/top-down inference)
Number of subimages	Precision	Recall	F-score	Precision	Recall	F-score
1	0.9262	0.9460	0.9359 ± 0.0074	0.9266	0.9498	0.9381 ± 0.0078
3	0.9378	0.9579	0.9477 ± 0.0068	0.9397	0.9580	0.9488 ± 0.0064
5	0.9495	0.9605	0.9549 ± 0.0065	0.9498	0.9599	0.9548 ± 0.0066
7	0.9510	0.9625	0.9567 ± 0.0081	0.9513	0.9620	0.9566 ± 0.0076
9	0.9526	0.9659	0.9592 ± 0.0052	0.9531	0.9661	0.9596 ± 0.0050