Noise Pre-Whitening

OpenMRF allows easy integration of noise pre-whitening scans into your sequence. In all example MRF sequences using the SPI readout module, this feature is enabled by default.

In the sequence creation script, add the noise prescans after initializing the SPI readout module and before adding any other objects to the sequence structure.

SPI.Nnoise = 16;     % number of prescans
SPI_add_prescans();  % script for adding prescans

In the reconstruction script, the acquired raw data are automatically separated into DATA and NOISE arrays:

% load MRF rawdata
[DATA, NOISE, PULSEQ, study_info] = pulseq_read_meas(path_raw_mrf, path_backup_mrf, vendor);

In the mrf_reco() function, noise pre-whitening is automatically performed if NOISE is not empty. By default, the whitening matrix is computed via Cholesky factorization:

% noise pre-whitening
if ~isempty(NOISE)
    DATA = mg_noise_prewhitening(DATA, NOISE(:,:,20:end), 'cholesky', 1);
end

Note

NOISE(:,:,20:end)
The first few samples are excluded from the whitening matrix calculation, as they may be affected by ADC filtering effects.
The value 20 is chosen empirically and typically provides sufficient padding for stable estimation.