Galaxy | Tool Preview

Perform wavelet transformation on 3D-image stack (version 0.1.2+galaxy3)
Select a movie to wavelet analyze
Preprocessings
Preprocessing 0
Wavelet Analyses
Wavelet Analysis 0
Create a mask from the (blurred/rescaled) input to mask out regions without oscillations/signal in the output

What it does

Wavelet transforms a 3d-image stack (time,Y,X) pixel by pixel along the time axis and extracts/evaluates the wavelet ridge. Removal of low-frequency trends is provided via sinc filtering. There are four output movies generated (phase, period, amplitude and power). They have exactly the same dimensions (time,Y,X) as the (rescaled) input.

Pre-smoothing of the data with Gaussian kernels is supported and often recommendable.

To limit the number of Wavelet transformations (computing time), downscaling of the input movie resolution is recommended. Trimming away parts of the movie with no relevant data (e.g. outer dark edges) also speeds up the processing.

If multiple channels are present in your data, extract the channel of interest beforehand (e.g. with Fiji).

Author: Gregor Mönke (github.com/tensionhead).

SpyBOAT repository: https://github.com/tensionhead/SpyBOAT

Wrapper by: Jelle Scholtalbers (jelle.scholtalbers@embl.de).

Know what you are doing

You need to have a basic understanding of time-frequency analysis with Wavelets, find more information about the analysis strategy employed at https://github.com/tensionhead/pyBOAT or the publication cited below.

Parameter List

  • Sigma:

    The Kernel bandwidth (in pixels) for the Gaussian kernels to use for pre-smoothing the input data. The default value of zero means that no pre-smoothing is done. Set this number to a desired kernel width to turn on pre-smoothing.

  • Rescale factor:

    Given in %, downsamples the spatial resolution of the input movie. This speeds up the whole analysis by a lot.

  • Masking:

    SpyBOAT offers 'static' and 'dynamic' masking. 'static' means that a single mask is created by thresholding from a user selected frame. 'dynamic' creates a mask for each frame by thresholding, works better for moving/expanding/shrinking oscillatory fields.

  • Sampling interval:

    Time passed between consecutive measurements, e.g 'an image every 10 minutes'.

  • Smallest Period:

    The minimal period to scan for, this is the higher (in frequency) end of the spectrum. A warning will be given during processing if the chosen value deceeds the Nyquist limit (2 times the sampling interval).

  • Largest period:

    The maximal period to scan for, this is the lower (in frequency) end of the spectrum. A warning will be given during processing if the chosen value exceeds the length of the time series.

  • Number of periods to scan for:

    This is the the number of convolutions computed per pixel.

    Spectral resolution = ( biggest period - smallest period ) / number of periods

  • Tcutoff:

    The cut-off period for the sinc filter, periods larger than this one will be removed from the signal before the transform. If no value is given, no detrending is performed. At least a gracious cut-off of around ~3 times the largest period to be expected is recommended.

  • Sliding window size

    Length (in time units, e.g. hours) of the sliding window to estimate an amplitude envelope. The signal then gets normalized by this envelope, meaning that all amplitudes will be around ~1 after the transform. The advantage is that signal with strong amplitude trends will have more meaningful Wavelet powers after normalization.