Advice on improving a analysis on images of nanoparticles

[Adam Hughes]

Hi Juan, thanks for your helpful response.  See my replies inline:


On Tue, Nov 19, 2013 at 9:39 PM, Juan Nunez-Iglesias <jni.soma at gmail.com>wrote:

> Adam,
>
> On Wed, Nov 20, 2013 at 7:44 AM, Adam Hughes <hughesadam87 at gmail.com>wrote:
>
>> To download without an account, I am not familiar with any hosting
>> solution, but if you guys have any recommendations I'd love to hear them.
>>
>
> Dropbox can be used for this...
>

I tried this but to share a link, it asks for the emails of the recipients.
 You have used Dropbox to host a publicly accessible link?  If so, I will
certainly start doing this, thanks.


>
> Thank you for you help, I will test out the methods you suggested
>>
>
> Is the goal only to count particles? In that case, I think a local
> thresholding (threshold_adaptive) would work on all these images. Then,
> just do a labelling (scipy.ndimage.label) and draw a histogram of particle
> sizes. You'll get a sharp peak around the true particle size, with bigger
> peaks for clumps. Once you have the mean particle size you can estimate the
> number of particles in each clump (barring occlusion in 3D, in which case
> you're stuffed anyway), and then the total number of particles in your
> image.
>

Yes, that is the goal.  We had done a similar process ImageJ, but did
thersholding manually.  I will read into the adaptive threshold a bit more.
 We had hoped that some of these corrections, such as histogram
equilization, would make the automatic threshold more likely to give
correct results.


>
> Looking at your images, I don't think watershed (or anything else that I
> know) will do very well with the clumps. The contrast between adjacent
> particles is too low.
>

Hmm I see.   I will still try it out, but thanks for the heads up.  I'll
feel better now if it doesn't work well.


>
> Low-contrast-4 looks tricky... Are the smaller "points" particles of
> different sizes or just image noise?
>
> Finally, Watershed-f3 also looks hard, because it appears all the
> particles are touching... Again, I don't think watershed will help you
> here, nor anything else that doesn't have an a-priori knowledge of the
> particle size.
>

We do have an a-prior knowledge actually.  What I've been doing already is
putting a lower limit on particle size, with anything under it being noise.
 After doing particle counts and binning the data, we fit it with a
guassian, and optionally scale the data so that the guassian is centered
around the mean partitcle diameter (which believe we know to about 3nm
based on TEM imaging and indirect spectroscopic techniques).  Based on the
size distribution, we try to further bin the data into small
(dimers/trimers) and large aggregates.  For all the particles that are
large enough to be considered an aggregate, we *assume *that they fill a
half-sphere volume, and then we infer the true particle due to these
aggregates.  It's pretty ad-hoc, but we certainly apply some knowledge of
the expected particle size distributions.  I realize watershedding won't
split up huge clumps, but maybe could assist in the dimers and trimers?  In
any case, even if it doesn't significantly enhance our results, it would
still be helpful to explore that option and I'll try it out.

Thanks for this, and other examples!


> Juan.
>
> --
> You received this message because you are subscribed to a topic in the
> Google Groups "scikit-image" group.
> To unsubscribe from this topic, visit
> https://groups.google.com/d/topic/scikit-image/EbS7o2HvcUc/unsubscribe.
> To unsubscribe from this group and all of its topics, send an email to
> scikit-image+unsubscribe at googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scikit-image/attachments/20131120/7633b297/attachment.html>