registration artifacts in 00 series due to line-by-line registration

[marius10p]

Based on my previous correspondence with Jeremy, I think the 00 datasets been registered with a line-by-line algorithm. Is it possible to redo this please?

It did not work very well. There are horizontal break points at specific Y positions in the image. Check the top 100 SVD components to see this. Not the very top ones, but everything after ~5 SVDs has horizontal artifacts. This happens for all datasets in the 00 series, and I get lots of ROIs that are just horizontal lines. I can still see the ROIs on top of these horizontal lines, but it's not ideal.

[sofroniewn]

@marius10p although the line-by-line motion registration generates those black line artifacts on individual frames, it may lead to a better signal to noise in the final extracted neuron traces due to its ability to better correct within frame motion and thereby lowering neuropil / other neuron contamination when compared with motion corrections done using a single global X and Y translation. Maybe the data should have been interpolated though instead of being blanked on the missing lines.

It would be hard for us as ask the original data provider to redo this data set for us now, and as there a number of groups using such line-by-line registration algorithms (which make a lot of theoretical sense for certain microscope designs / line rates / frame rates) I think it will be useful for the community to see how the different neuron finding algorithms perform on this type of dataset. If they all perform poorly it could argue against performing motion registration in this way.

[marius10p]

Do you have a reference for this comparison? I would be curious to see if it indeed performs better, I have never tried it myself. Another argument is that there is too little SNR to do line-by-line registration anyway, and the data just gets grossly corrupted in this fashion. In fact, the SVD decomposition seems to argue that the signal variability created by this procedure is greater than most of the neural signals. Maybe the interpolation would have helped.

[sofroniewn]

Unfortunately those comparisons were never published, but it was something looked into by Simon Peron maybe 5 or 6 years ago in Karel's lab. I think it was particularly important for the older galvo-galvo (i.e. non-resonant scanner) microscopes. I'm now using a global registration approach which doesn't have those artifacts. Maybe we should set up a website with standardized data sets and ground truth to compare motion registration algorithms ;-)