Neurons
From SeedWiki
Detection of Neuronal Somas
The project will develop methods for segmentation of neuronal somas for automated analysis of large datasets. We also plan to use boosting for generating region of interest masks for intelligent scanning with TPLSM.
Ilya Valmianski 17:32, 28 July 2010 (UTC)
Here is a link to the finished article. [1]
--Ilya Valmianski 18:06, 12 December 2008 (UTC): Ok, in an effort to bring up to date the FTP server and all that has been done, I'm uploading all of Andy's data. I will also soon upload all of my work on it. Some trials have two channels, "Ch1" is the Ca2+ data while "Ch2" has fluorescent indicator of astrocytes.
--Ilya Valmianski 18:02, 4 December 2008 (UTC): Here's a link to wiener filter: [2] . In MATLAB denoising using wiener filter is done with wiener2, deblurring using wiener filter is done through deconvwnr. I think both are in Image Processing Toolbox.
A lot of updates on my diary page... --Ilya Valmianski 06:41, 27 November 2008 (UTC) Edit: Explanation of my image processing algorithm can now be found in Regional Automatic Gain Control.
I've uploaded to the ftp images representative of the preprocessing that I have done. I will assemble them into a power point before the meeting with DK on Tuesday. --Ilya Valmianski 04:13, 18 November 2008 (UTC)
Variance, covariance, correlation, pca, average, svd, gradient images for one set of Andy's data are all online. --Ilya Valmianski 01:14, 15 November 2008 (UTC)
Updated covariance image processing so that window moves by 1 px instead of entire window size. New data in "better covariance data + m files.zip" --Ilya Valmianski 02:03, 13 November 2008 (UTC)
Hello Yoav,
I've put on ftp two of Mariano's datasets into the "For Yoav" directory. The folders are "Mariano data1" and "Mariano data2". I've done edge detection of the first dataset ("Mariano data1") and the results are also in the "For Yoav" directory. The four types of detection I did were:
1) Average image with automatic gain control ("agcimage.tif")
2) First 10 modes of SVD ("svdimage.tif")
3) First 2 components from PCA ("pcaimage.tif", really, the first component has nearly all of the data, it's 8 orders of magnitude more powerful than the second component)
4) Covariance calculations. there are a total of 12 covariance images, and they are labeled as cov_CORDINATE_WINDOWSIZE_image.tif . So, covx1image.tif means I computed covariance along the x axis and window size is 1x1.
Also, there is a mat file "images.mat" which has the original 256x400 matrices that are the images (they are named same as the tif files).
I think that the first covariance edge detection is very good for window sizes of 1x1 and 2x2, but not larger. SVD/PCA/average also give good pictures.
--Ilya Valmianski 22:31, 12 November 2008 (UTC)
hi Yoav,
this makes sense and is straightforward to implement; i will try symmetric conditions first (i.e. what you call sizes 1 and 3 below). two other things:
1. we should coordinate how we deal with edge effects, i.e. computations on pixels near the border of an image. i have preferences for dealing with this issue, but we should make sure to be clear about each step of the algorithm.
Yoav: As long as the templates are small relative to the size of the image, edge effects are negligible. I suggest you document your preference here, just for the record.
2. is there any help file for the matlab code? i read through the text file included with the software, but there must be some master '.m' file for running each function in sequence on a particular data set. please advise.
Yoav: It is a bit more complex than that. The "master" file is a python file, not a .m file, there is no documentation. You'll have to talk with Mayank and have him teach youhow the code is organized. Come to think of it, it would be a great contribution is you would then write a help file of the type that would help you right now.
dave: i'm willing to write a help file. i'll coordinate with mayank to learn the .py code. thanks.
Hi David,
I thought some more about how to make the best use of the correlation-across-time information to identify the pixels that correspond to a single cell. I think that instead of calculating the average correlation between a pixel and it's neghbors
Edge Features --Yoavfreund 04:48, 22 October 2008 (UTC)
