Computer detection of the rapid diffusion of fluorescent membrane fusion markers in images observed with video microscopy.

We have developed an algorithm for automated detection of the dynamic pattern characterizing flashes of fluorescence in video images of membrane fusion. The algorithm detects the spatially localized, transient increases and decreases in brightness that result from the dequenching of fluorescent dye in phospholipid vesicles or lipid-enveloped virions fusing with a planar membrane. The flash is identified in video images by its nonzero time derivative and the symmetry of its spatial profile. Differentiation is implemented by forward and backward subtractions of video frames. The algorithm groups spatially connected pixels brighter than a user-specified threshold into distinct objects in forward- and backward-differentiated images. Objects are classified as either flashes or noise particles by comparing the symmetries of matched forward and backward difference profiles and then by tracking each profile in successive difference images. The number of flashes identified depends on the brightness threshold, the size of the convolution kernel used to filter the image, and the time difference between the subtracted video frames. When these parameters are changed so that the algorithm identifies an increasing percentage of the flashes recognized by eye, an increasing number of noise objects are mistakenly identified as flashes. These mistaken flashes can be eliminated by a human observer. The algorithm considerably shortens the time needed to analyze video data. Tested extensively with phospholipid vesicle and virion fusion with planar membranes, our implementation of the algorithm accurately determined the rate of fusion of influenza virions labeled with the lipophilic dye octadecylrhodamine (R18).