Phase regression

[tsalo]

Look at https://github.com/brettliem/phaseprep and https://github.com/idevicente/Phase-based_denoising_pipeline

[tsalo]

Steps:

1. Remove non-steady-state volumes from phase data (identified from magnitude processing).
2. Compute optimal order of Legendre polynomials.
3. Convert phase from arbitrary units to radians.
4. Combine magnitude and phase into complex-valued data.
5. Split complex-valued data into real and imaginary components.
6. Apply motion correction transform from magnitude processing to real and imaginary files.
7. Combine motion-corrected real and imaginary files into complex-valued data.
8. Split motion-corrected complex-valued file into magnitude and phase components.
9. Temporally unwrap the motion-corrected phase data.
10. Detrend the unwrapped phase data using the Legendre polynomial order determined previously.
11. Apply brain mask from magnitude processing to phase data.
12. ODR phase regression.
    1. Compute the power spectrum map of the phase data
    2. Find the time point index of the maximum frequency peak starting from the 93th time point.
    3. Compute the value that happens the most in the previous map to find the time point index of the respiration peak.
    4. Multiply this index by 1.1 to consider the subpeaks around the maximum peak.
    5. Convert it to frequencies (Hz) by multiplying it with the frequency value of the first time point.
    6. Run the ODR regression.
13. Run standard GLM on phase-regressed magnitude data.
    • I'm not sure why the GLM is done separately from the phase regression. Maybe because the phase regression requires ODR?