CellSegm - a MATLAB toolbox for high-throughput 3D cell segmentation

Table 3 Quantitative comparison of segmentation accuracy between CELLPROFILER and CELLSEGM as measured by falsely fused, splitted, positive and negative cells for all four data sets

Comparison/	Falsely fused	Falsely splitted	False positive	False negative
O₁-CellSegm	15	13	1	1
O₂-CellSegm	6	11	2	8
O₁-CellProfiler	72	4	0	49
O₂-CellProfiler	62	0	0	56

CELLPROFILER has much more falsely fused cells and false negative cells than CELLSEGM, although CELLSEGM has more falsely splitted cells. The latter is mostly due to cells having two nuclei, which is not uncommon for cancer cell lines. The false positive and false negative rates are very low for CELLSEGM. The robustness of CELLSEGM is greatly improved when the nucleus approach is applied, where a cell is initiated only if a valid nucleus marker exists.

ISSN: 1751-0473