This dataset contains time series of reconstructed image stacks, one set per tablet composition. The reconstructed image stacks were binned from their original resolution to suit the memory constraints of our deep learning image segmentation algorithm. Each time series of reconstructed image stack is supplemented by its equivalent time series of segmented image data. The dataset additionally contains the hand-segmented data that was used to train the CNN.
This dataset is structured as follows:
For each tablet formulation (N1 to N64), a folder is available containing the full time series of reconstructed image stacks (1 HDF5 file per time point). A second folder is available for each formulation containing the same series of image stacks in segmented form (1 HDF5 file per time point). The trained U-Nets used for the segmentation of the data are available in a separate folder, along with the hand-segmented data used to train these networks. Lastly there is a ‘readme’ folder containing documents detailing acquisition parameters as well as details on the composition of the different tablets.