Best approach to threshold growing comet like structure

Hello everybody,
in most of my experiments I follow overtime the accrual of an insoluble protein over a coverslip using a monoclonal antibody coupled with a fluorochrome, with epifluorescence microscopy. To give more context this is done in flowing whole blood, the protein is normally soluble, but in the right conditions it is not and “precipitate” around cells and create a comet like fiber. This accrual happens only after a couple of minutes from the beginning of the experiment. So far, we have had fair success using a Niblack algorithm. More recently I’m trying to use an adaptive threshold algorithm from the OpenCV library. With both algorithms though, I end up having a less than optimal situation, because the parameters that work well for let’s say a later time point in which the fibers are big, do work poorly when there are no fibers at the beginning of the experiment, normally overestimating the signal. Right now, at least with the Niblack we created 2 masks and we put them together, but we want to make this analysis automated so I would prefer to be able to find a way to make it work automatically. I’m super interested in understanding in general, which approach to segmentation I could take, even different than thresholding. Also, I wonder if there is any transformation that could be done to make the image analysis easier. I’m attaching a couple of frames taken at two different time points: left early/ right late time point. If anybody is up for the challenge, I’d be happy to discuss further!