Single Particle Tracking

SMTrack package implements a robust single molecule tracking scheme based upon a modified version of Lineage Mapper (Chalfoun et al, Scientific Reports 2016). A novel implementation of a per-frame particle jump-distance has been incorporated, which relies on a mean separation distance (computed via a Delaunay Mesh) and its minimization such that a maximum of 0 or 1 particle-pair association is obtained between consecutive frames. There is also the option to specify a search distance. A graph theoretic approach globally optimizes the cost matrix to yield appropriate linkages.

The implementation is expected to work successfully with numerous particle detection strategies. Another aspect is the inclusion of "subpixel particle localization" which relies on a two-dimensional Gaussian-Fit.

Note: the current detection scheme utilizes a Laplacian of Gaussian (LoG) filter with thresholding. Another variant uses the maxima of the distance transform of the thresholded LoG-filtered image as seeds to WatershedComponents (gradient descent version or rainfall algorithm (Osma-Ruiz)) to segment for individual spots. 2D Gaussians can be fit to either the thresholded image or the watershed components.

Possible improvements (Near Future):

1. Add a different detection scheme (example, FogBank by NIST)

2. Not necessary but perhaps important for low density cases: Kalman predictor to handle occlusions when a particle motion-model exists (e.g. constant velocity etc ..)

Input:

for computing per-frame jump distance:

Note: I have not updated the image below but the improved detection scheme in the code captures almost all visible particles

superposition of all particle trajectories for the above movie subpixel localization = False:

a Gaussian (rainbow) is fit to the data (yellow) for achieving subpixel localization:

superposition of all particle trajectories for the above movie subpixel localization = True: