The Laplacian-based algorithm was also tested on six datasets from the ISBI 2013 Cell Tracking Challenge and six datasets from the Cell Tracking Challenge 2017. For the first six datasets, we evaluated the segmentation accuracy using the Jaccard similarity index. Our algorithm was seen to perform the best on three of the datasets and produced comparable results on the other three as well. For the last six datasets, we carefully picked unique types of cells and cell images to determine the strengths and weaknesses of our algorithm. Our results on the Fluo-C2DL-MSC and Fluo-N2DH-GOWT1 datasets demonstrate that the algorithm is able to effectively segment irregularly shaped cells as well as extremely low intensity cells. Furthermore, we showed that the algorithm is faster and better able to segment 3-D images as a whole than segmenting each slice separately.
SCHEMELECT Full 16
2ff7e9595c
Comments