At the ISREC, studies on neuron development are conducted that requires biologists to perform morphological measurement (such as length of dendritic tree, number of branches, angles between branches) on neurons images obtained via confocal microscopy. Our goal is to make these measurements easier.
The aim of the project presented hereafter is to design an ImageJ plug-in allowing automated tracing of neurons from confocal microscopy image stacks. This task is currently almost done by hand, and is extremely time-consuming and error-prone. The developed plug-in would allow almost full automation of the process, and tracing in 3 dimensions.
To obtain the desired tracing, we used several methods. The first method we designed uses :
a 3D steerable filtering to enhance the initial stack and obtain the orientations of the pixels;
a threshold to segment a search space from the resulting stack;
Dijkstra to compute a shortest paths tree covering the segmented search space;
a tree pruning algorithm to isolate the relevant branches of the tree;
manual editing of the result;
automatic measurements on the resulting tree;
The second method designed is identical except that it uses a hysteresis threshold instead of a threshold.
We also tried a third method, with Dijkstra applied to the whole stack, but it did not yield usable results.
We were able to implement a working plug-in that allows tracing of neurons in a lot less time than allowed previous methods. Improvements are however still needed to improve the usability of the plug-in.