Additional file 3 of ClearFinder: a Python GUI for annotating cells in cleared mouse brain

Supplementary material 3: Figure 3. Image cubes for cell detection performance assessment. Examples of image cubes of sample #1 (A), #2 (B) and #3 (C). White dots mark counted cells of expert 1, red dots of expert 2, yellow circles of ClearMap and green circles of CellFinder. For sample collection 18 randomly selected x,y and z coordinates were generated. With these start coordinates cubes with a volume of 60µm3 of the raw signal files were created using ImageJ. Due to inevitable resizing events during the Cell detection process with CellFinder, the coordinates had to be re-transformed to the original signal image sizes. The latter was performed by multiplying the determined coordinates with the relative size difference of respective dimensions. Subsequently, the determined coordinates of both sub-packages were transformed to fit the coordinates of the extracted cubes. This could be achieved using the 18 selected x,y, and z coordinates as subtrahends for the determined coordinates of both sub-packages. All coordinates within the boundaries of the 60µm3 cubes with a tolerance of ± 5µm were selected for cell detection assessment and stored in xml files. The image cubes were visualized in Napari with the option to switch between 2-dimensional (2D) stacks and 3-dimensional (3D) visualization. Two human experts were counting cells in 2D stacks first. Subsequently the experts switched to 3D space to filter for redundantly counted cells in the previous step. After the human experts finished counting the transformed coordinates of ClearFinder’s sub-packages were opened in Napari and the cell counting was assessed in 3D space