Notes:
How does it work?
Imagine the image as a landscape where light regions are mountains. In the landscape neurites are ridges.
In a preprocessing step a second order differential operator (hessian) is used to get the directions of the ridges and to calculate a likeliness for each pixel to belong to a neurite.
When a tracing is started, the cheapest path from the start point to the mouse pointer is calculated using the data from the preprocessing step.
The red marks in the image reflect the directions of the ridges.