Contains code and script for data analysis (exploration and preparation) for VACV Plaques dataset.
These options are used with trackmate to generate spots data. Trackmate batcher with an example xml file can be used to parse batch of tiff files. It will create a .csv file for each .tiff file. The original analysis provided with the dataset were not used, because they lead to incorrect spot locations as can be seen in here. If you check for different files, they lead to different offsets which are not consistent. Therefore I ran the analysis again using trackmate and the config provided.
Each .tiff image in original dataset has two channels, where one of the channles is useless (some contain info, but most seem to be just noise). Therefore only first channel is kept using fiji and this script.
This is used to look at the results of trackmate, and also to copy csv files to where the .tiff files are to create a dataset.
original .tiff files are in the format uint16, which is not supported by pytorch (ToTensor() transformation in torchvision does not support uint16). Also, these files have a range much smaller than the dtype uint16 allows. Therefore using cv2 to linearly rescale the values to the range of uint8 dtype. max and min values for rescaling is computed using each .tiff file (contain 169 frames) and not the whole dataset. This notebook does this.
create reference metrics notebook is used to create a reference metrics csv file to evaluate simulation results. Current metrics used are: mean and std values of infected cell counts, and mean and std of the radius of the plaque at each time of experiment hours post infection. and radial velocity. Also added another column to each metric, a list of all the numbers that mean and std were computed, such that 2sample tests can be used.
Currently not using all the available dataset, but only the ones that are clean. That is the ones that have a single plaque center.
motility analysis notebook shows frame to frame (backward difference) speed for different strains with inf_time (frames passed after infection) as parameter.
direction deviation angle notebook uses trackamte results to find the angle between the frame movement direction and a central moving average of the movement over a longer period. Was hoping for smaller angles but std of the angles are around 60 degrees which is too large to be consired as just noise I would say.