Local maxima in this parameter space can be thought of as centroids of cells. This strategy is beneficial for detecting cells with low-contrast boundaries due to the ability of the CHT to detect shapes based on non-contiguous and partial set of edges. Furthermore, it bypasses the need for segmentation LEE011 purchase of individual cells and thus aid in

the accuracy of detection in high-density environments (Fig. S1 for example). We have used Tao Peng’s implementation of the CHT (CircularHough_Grd from the MATLAB File Exchange repository) as it considers a radius range during the voting process and includes an additional parameter for searching maxima over imperfect circular shapes. Accordingly, we have found our implementation to detect polarized T cells as well as cells of different types, morphologies and at different cellular densities in images acquired by all three aforementioned transmitted find more light microscopy techniques

(Fig. 2, Fig. S1, Fig. S2, and Videos S1 and S2 and Video S3). The individual parameters involved in the detection step are described further in the Supplementary methods section. Parameter values typically used in our T cell imaging experiments are also provided. Successful detection is critical for all the ensuing computational steps. Therefore we have developed a graphic user interface in Java to interactively change parameters of the Canny-edge filter and CHT to achieve successful detection of cells in transmitted light images. The user guide provides an example of this process to help with intuitive selection of parameter values. The user is prompted to adjust the scale of the image such that the cell size is similar to the example provided in the user guide. This attempts to ensure that the default radius range used during CHT voting process works well. Similarly, edge detection and additional CHT parameters can

be chosen by comparison to the example images of these stages. The centroid positions are transformed back through to the original scale at the end of the detection step, before proceeding with tracking cells. Tracking in TIAM is carried out in two steps. In the first step, a modified nearest neighbor association algorithm is applied to the outputs of the cell detection step to yield short track ‘segments’ (Fig. S3a). At each time step t, each cell is linked to the spatially nearest detected cell of the previous time step t − 1, provided the nearest detected cell is within a maximal allowed distance r. This process proceeds in this manner only when cells are sufficiently separated and there is no tracking ambiguity. If there is more than one cell within r, the algorithm returns the track segment that has been produced up to that frame and initiates new tracks with neighboring cells that caused the ambiguity.