ABSTRACT
The combination of fluorescent probes with time-lapse microscopy allows for the visualization of the entire neuronal activity of small animals, such as worms or cnidarians, over a long period of time. However, large deformations of the animal combined with the natural intermittency of neuronal activity make robust automated tracking of firing fluorescent neurons challenging. Here we present an hybrid approach where (i) a subset of very bright neurons is used as moving reference points (fiducials) to estimate the elastic deformation of the animal; (ii) deformation is frame-by-frame corrected, and firing neurons are aligned at each time with the initial mask; and (iii) point-set registration is used to robustly track the intermittent activity of all the immobilized neurons. We compare different registration strategies with manual tracking performed over ≈620 neurons over 100 time frames in the cnidarian Hydra vulgaris.
Index Terms Fluorescence imaging, wavelet detection, tracking, point-set registration, elastic deformation, Thin Plate Spline (TPS) transform, Coherent Point Drift (CPD), Hydra.