Data pre-processing

fUS data must be pre-processed to remove the effects of motion and noise as to limit the number of artifacts in the later analyses. Motion correction requires both correcting the lateral motion (shifts up and down or left and right) of the image due to the movement of the probe relative to the brain, as well as the small number of “burst” artifacts caused by extreme vigor of the animal. The latter “burst artifacts” are easily detectable by their global increase in image intensity, as in this figure

but cannot be easily corrected and thus those frames must be removed and replaced, e.g. via in-painting. Lateral motion can be corrected with rigid or nonrigid frame-wise alignment. We find that simple rigid correction is sufficient to align even data from different days. As an example, we can plot the offsets estimated via cross-correlation functionsacross the recording duration, and match that with fluctuations at the single pixel level:

To assess the extent of motion correction, we test how different the per-pixel distribution of values are during high motion times from low motion times via a histogram-based metric:

which can be plotted across the imaged area to see which areas are most impacted by motion

Denoising is helpful to isolate the signal for more accurate analyses. We find that 1D wavelet denoising performed per-pixel removes most noise while not blurring the spatial image. Other methods, e.g., averaging, deep learning, etc. are also possible and may have different strengths and weaknesses.