Comparison with Alternative

After tests on over 150 neuron images, we are confident to say that our program is better than a similar program developed by a Multidisciplinay Design Program (MDP) group from University of Michigan (Angileri et al 2013). We categorize the sample images into “fine images” and “problematic images” depend on their quality and the results of previous MDP program. Both our program and previous program perform well on “fine images”. However, our program does pretty well on most of problematic images while previous program perform poorly. In general, tracing results of our program is clearer than ones of previous program.

Performance on fine images

As we compare the tracing results of our program with ones of previous program, we find both of them perform well. Here are some examples.

Performance on problematic images

The problematic images are those who have a lot of noises and bright background that it’s quite hard to detect edges. However, our program perform well on those images compared to the previous program, because their program is sensitive to over-brightness and will generate unwanted cellular patterns. The following are some examples.

Limitations

There are still some images hard for our program to trace, although the outcome of our program is better than the existing program. Sometimes, because of aliasing between over-bright background and neuron as well as intensive noises, our tracing lose some part of neuron. The following are two examples.