Difference between revisions of "Data Processing"

From UCLA Miniscope
Jump to: navigation, search
(fixed link)
 
Line 9: Line 9:
 
[[File:MINIscopeTraces.png|650px]]
 
[[File:MINIscopeTraces.png|650px]]
  
Download our MATLAB analysis package here: [[File:Analysis_Package.zip]]
+
Download our MATLAB analysis package here: https://github.com/daharoni/Miniscope_Analysis

Latest revision as of 10:56, 4 December 2017

Data is recorded by our DAQ software in uncompressed .avi format and linked to time stamp information. Our analysis suite, written in Matlab, processes this raw data and extracts relevant experimental information. The analysis is built around a custom data structure which stores location and processed manipulations of the raw data without storing the raw data itself. This results in very small files sizes while still maintaining quick access to data and previous analysis.

The initial processing of microscope data involves correcting for column wise ADC variation, image registration of each frame using an amplitude based algorithm, and conversion to dF/F. At this point our novel automated segmentation algorithm segments pixels of active cells. A brief description of our segmentation algorithm is it runs through the dF/F video looking for local bright spots, When a bright spot is found, the algorithm runs an iterative process calculating correlation coefficients (in a ~+/-5s window) of surrounding pixels with the center pixels of the bright spot. Pixels with high correlation get grouped into the segment and the process runs again until some criteria is met. Once segmented we extract dF/F traces (Figure 8), spike times, and activity measures. With recording from an animal across multiple sessions the analysis suite aligns and matches segments segments using distance and overlap measures.

An issue with our imaging data, and wide-field imaging data in general, is fluorescence cross-talk between neighboring cells due to scattering of emission light. We are in the process of developing novel techniques to remove or minimize this cross-talk using approaches like blind source separation and spatial deconvolution. While these novel techniques are still a work in progress, we do have another approach to removing cross-talk which silences the neighboring segments around an active cell. This approach works reasonably well but removes a nontrivial amount of otherwise useable data.

Along with processing microscope data, the analysis suite tracks animal movement in behavioral recordings and extracts position and velocity information. Using this information we can calculate place field information.

MINIscopeTraces.png

Download our MATLAB analysis package here: https://github.com/daharoni/Miniscope_Analysis