Figure 2

Manual examination of segmentation results. Manual examination of segmentation results, shown at exemplary image patches over the whole time span of a 6 day time-lapse experiment of differentiating hematopoietic stem cells. Blue outlines: Segmented objects regarded as cells. Red outlines: Objects unlikely to represent cells (size <50 px and eccentricity >0.99). First row: 500x500px image patch, second row: 150x150px image patch. (A) 12 hours after experiment start. Very few cells are populating the field of view. Cell outlines are correctly segmented. Erroneous measurements originate in debris in the image. (B) 2 days after experiment start. The number of cells is slightly increased, still the object density is very sparse. Pairs of clumped cells can be identified, which are correctly split by the method. (C) 4.5 days after experiment start. More complex cell morphologies arise that lead to errors in segmentation. The field of view becomes more and more crowded, complicating the identification of single cells. Small artifacts that are a result of over-segmented cells or fragments of dead cells are filtered by size. (D) 5.5 days after experiment start. Most cells are differentiated and different morphologies can be found. Segmentation errors are observed more frequently, especially for adherent cells with elongated shape.