10um/pixel reference atlas

[ghost]

Hi there-

I'm wondering if there is any way to preserve image resolution by avoiding the step of downsampling slice images to 800 x 1400 pixels. I have a group of images that average a resolution of ~0.5 um/pixel and the loss of image quality upon transforming to reference atlas images makes for much data loss in cell activity. Would it be possible to upsample a specific atlas section by 20x once matched with a slice before transforming and overlaying the images?

Thanks

[philshams]

Hi, thanks for your question -

A number of solutions are possible, but it depends on exactly what you need to do. In theory, if you don't run into memory issues, you could upsample the entire atlas volume and atlas-labels volume 20x (it will be blurry but that should be ok), and then perform registration. But I wouldn't recommend this unless other solutions do not work.

The more straightforward solution is to adapt what's mentioned in the wiki with regard to ROI detection. There are two approaches listed there - transforming your 'activity' image itself during registration, or registering your slice image and then transforming the downsampled activity image post-hoc. This of course requires that your downsampled 'activity' image contains the same information as the full-size one. It's unclear why you have "much data loss in cell activity," since I don't know the details of your case. If activity is measured by the sum of pixel intensity in an ROI and all ROIs are separable even after downsampling, then you should be able to downsample without losing any information about cell activity. If the ROIs merge into each other during downsampling, you will indeed need an alternative solution.