*Segmentation related issues* /1. Tracker appears to have been trained on reasonably round cells e.g. HeLa. Would be useful to be able to alter boundary parameters to allow better tracking of more convoluted edges with lots of protrusions for some cell types compared to the more rounded ones. Suggested ranges and explanations for any parameters needed to manage this would be helpful./ The active contour method segment the image via minimising an energy function which is related to the curvature of the boundary. The round cells will have lower energy and is preferred by the algorithm. On the other hand, the method also incorporates the intensity information into the energy function. In other word, it depends on the contrast of cells with more convoluted edges and lots of protrusions. If the contrast is high enough, it should be segmented properly. The problem is that for most of the cell types the contrast is low, which leading to the segmentation in round shape. In theory, we can decrease the weight of curvature related energy term to allow edges to be more convoluted, however, other issues such as boundary might cross itself might arise. /2. Boundary shouldn't be allowed to cross from dark to light to dark again or vice-versa/ It is difficult to define what is "dark" and "light", since the intensities of nucleus and cytoplasm change dynamically. In some cases (e. g. the protrusion region), it is actually allowed to cross from dark to light to dark again. /3. Recognise contours rather than just threshold, might be helped by using edge detection tool/ The output boundary is based on the segmentation results, not just threshold the image. Segmentation by using edge detection tool is only suitable when the image intensity are quite homogeneous. For cell images, many edges will be detected due to the noise, which is hard to choose the right ones and connect them together to form a closed boundary. /4. Penalty for deviating from ellipse/diamond shape might help draw cytosol boundaries sensibly in low-intensity frames./ You are certainly right, it will help draw cytoplasm boundaries sensibly in low-intensity frames by adding a penalty term for deviating from ellipse/diamond shape. A segmentation method with ellipse shape prior should be developed. /5. Force boundaries between two cells not to cross a user-drawn line/ The user-drawn line can be consider as hard constraints provided by the user. By enforcing such kind of hard constraint into the energy function, the two cells will not allowed to cross the user-drawn line. * * *Tracking related issues* /1. Boundaries must stay within frame even if cell moves out of view./ This is a problem on how to deal with the situation when the tracking is lost. We can put a fake boundary within the frame by assuming there is no motion, which will give wrong information in most of the cases. Sometimes, it is quite difficult to tell whether the tracking is lost or not due to the similar appearance of different cells or the changing of appearance. /2. Might be helped by using track as ellipse then refine edges function, but can't get snake parameters on refine active contours to do anything useful/ The current method use the boundary in the previous frame as the initialization of the active contour and then refine the contour, which is more reasonable for most of the cases. When the segmentation going wrong, it will be better to use ellipse and then refine it. The question is where we should put the ellipse when the tracking is wrong. As in most cases, the wrong tracking is caused by the low intensity of cell, and it will not work well by just changing the parameters to refine the contour. /3. If using key templates tracking function would be useful to be able to detect cells in all key frames in one go using same parameters rather than each individually/ We can do this by combine the key frames into a 3D stack and segment it in one go, however the accuracy will not be guaranteed, especially when the cell changes dramatically in time. /4. Use threshold between values rather than simply above threshold for tracking cytoplasm/ The tracking of cytoplasm is not based on refining the active contour initialized by previous frame, not on the threshold. *Topology related issues* /1. Nucleus boundary must stay within cytoplasm boundary./ In the current implementation, the nucleus and cytoplasm are segmented separately. To tackle this problem, we are going to develop a method which preserves the topology of nucleus and cytoplasm by integrating knowledge about topological properties. /2. Nucleus boundary must not overlap cytosol boundary, one boundary must not overlap another/ Introducing the inter-repelling mechanism and enforcing the topology constraints should solve the problem. /3. Boundary shouldn't be able to cross itself/ Again, this is a topology problem, currently not implemented. *Cell type related issues* /1. Can you save a set of parameters for each cell type rather than having to alter them all by hand if you move from one cell type to another?/ We can tune the set of parameters to optimise the performance according to the specific type of cells, and save them into a configuration file. Every time before running the algorithm, you need to load the configuration file first. We need at least 3 data sets for each type of cell to tune the parameter. If you think your data is different from others, please upload 3 data sets to our server. /2. Need suggested range for all parameters and whether linear/logarithmic, need suggested values for each cell type to improve tracking/ The parameters are depending on the image data. We will test the different type of cell images and provide a suggested range for all parameters, and a configuration file for each type of cell. /3. Maintain set parameters rather than return to default each time, would be even better to be able to save sets of parameters for tracking e.g. if tracking nuclei separately from cytoplasms/ See above.