A wellstudied model for cell motility leading to metastasis includes Met tyrosine kinase receptor and its ligand, Hepatocyte Growth Factor/Scatter Factor. A better understanding of the changes that occur during HGF/SF-induced motility and development of new anti-metastatic targeted therapy are considered major challenges in biomedical research. Here, we investigate HGF/SF-induced cell motility via a novel approach that is based on Machine-learning classification to segment and analyze cellular regions in bright field images, similar to the general framework described by Shamir et al.. To compare the MultiCellSeg with alternative approaches, we considered the available automatic tools for wound healing analysis. Several researchers use a combination of edge-detection or simple local texture descriptors and morphological operators. These tools can be tuned to fit specific data sets but bear difficulties in handling diverse ranges of image-acquisition conditions and different cell types. CellProfiler has many useful applications, but its wound healing algorithm is using generic modules that are more appropriate for other applications; its performance in segmenting wound healing images under the default settings is very poor hence direct comparison was discarded. To the best of our knowledge, the only freely available software for automatic analysis of wound healing that performs reasonably well on bright field images without specific parameter setting is TScratch. The quality of MultiCellSeg was therefore compared with it. Both MultiCellSeg as well as TScratch can be seen as composed of two parts. First, the original image is used to create a new one, in which the intensity of each pixel represents the algorithm’s confidence in its classification. Then, this image is used to define the final ROI. The first phase in TScratch is the construction of the curvelet magnitude image, whereas in our approach, it is the generation of the classifier’s confidence image. The second phase in TScratch is the automatic setting of a threshold and then the application of morphological operators. In MultiCellSeg, the second phase includes removal of erroneous tagged regions and contour refinement. Thus, the comparison of these algorithms is performed in two steps. The robustness of the first phase is measured by examining the Receiver Operating Characteristic which plots truepositive versus false-positive classification rates of the pixels in each image across the entire range of possible thresholds of the confidence threshold, encoding the true potential of the underlying approach. The second measure is a direct comparison between the algorithms’ final tagging. The ROC SCH727965 curves comparing TScratch with MultiCellSeg are presented in Fig. 3. The x-coordinate represents the false-positive rate, which is the percent of pixels that were incorrectly tagged as background, out of all cellular pixels of the given image. The ycoordinate is the true-positive rate, which is the percent of background pixels correctly tagged, out of all image’s background pixels.