automated neurite tracing macro

[Madeline Pool]

Hello all,

I've been working on a macro for ImageJ for automated neurite  
measurement in neuron outgrowth assays. Our outgrowth assays are done  
with neuronal cultures that are immunostained and imaged by  
epifluorescence microscopy. For any interested non-biologists, we end  
up with greyscale images of bright neurons on a black background and  
we need to measure the lengths of the snake-like extensions from the  
ball-shaped cell bodies of the neurons.

I'm not a programmer so for the most part I have been looking at the  
code in other macros and recording commands to put a macro together  
to automate the tracings. I have a working macro that essentially  
processes the images to clean them up a bit then uses the skeletonize  
command to obtain traces. It does a few more things along the way as  
well like counting the neurons and measuring the traces to make a  
results table with an entry for each processed image. I've compared  
the results of my macro to those obtained by tracing with NeuronJ and  
they are usually within 10%. This is good enough for our lab's high  
throughput screening purposes as we are not interested in subtle  
effects. The macro's main problems are that 1) thresholds are set  
manually for each batch of images so there is fairly high potential  
for variability 2) not all neurons are traced because fainter ones  
are ignored and 3) dirty images with speckles, bubbles or junk give  
false positive traces. These are not major problems for us because 1)  
we are always comparing to control untreated cultures in the same  
batch of images, 2) we don't expect any of our treatments to change  
the proportion of faint neurites (this is always verified visually as  
well) and 3) our images do not contain many speckles and if there is  
a bubble in the way we can always take another photo of a different  
region of the culture.

I have looked at several papers describing algorithms for neurite  
tracing. These are all far more sophisticated methods than mine and  
yield more accurate results. However, they are usually written for  
use with Matlab or other platforms and so are not easily accessible  
to the majority of biologists. NeuronJ is accessible however it  
requires selecting beginning and end-points for each trace and is  
therefore still very time-consuming. So, I'd like advice on whether  
it would be worth writing a short technical article describing my  
macro and which journals might be interested in publishing such an  
article (Cytometry and maybe Journal of Neuroscience Methods are the  
two main ones that I can think of).

Finally, I don't know Java and have looked up the ImageJ Plugin  
writing tutorial but find it intimidating. Would it be better to  
convert the macro to a plugin or is a macro just as good?

Thanks for any input,
Madeline

Madeline Pool, PhD
Department of Neurology and Neurosurgery
Montreal Neurological Institute, BT-105
Montreal, Quebec
H3A2B4
TEL:  [phone]
FAX:  [phone]