nifreeze.testing.phantoms module

Noise of the synthetic reference datasets (see Synthetic reference data).

The simulator writes noise-free images. Noise is added afterwards, so that one noise-free volume can be stored once and given the same noise in every dataset that contains it, perturbed or not: the noise of a volume is drawn from a generator keyed on a seed and on the volume’s index in the full dataset.

nifreeze.testing.phantoms.add_noise(volumes: ndarray, sigma: float, seed: int, indices: Iterable[int] | None = None) → ndarray

Add complex Gaussian noise to noise-free volumes, and take the magnitude.

The noise-free volumes are real (their phase is zero). Each channel gets noise of standard deviation sigma, so the magnitude follows a Rice distribution, and a Rayleigh one where there is no signal.

Parameters:

• volumes (ndarray) – Noise-free volumes, of shape (N, X, Y, Z).

• sigma (float) – Standard deviation of the real and of the imaginary noise.

• seed (int) – Seed of the noise realization.

• indices (Iterable, optional) – Index of each volume in the full dataset (0, 1, ... by default), which keys the noise of that volume.

Returns:

The noisy magnitudes, float32, of the same shape.

Return type:

ndarray

Examples

>>> volumes = np.zeros((3, 4, 4, 4))
>>> noisy = add_noise(volumes, 1.0, seed=7)
>>> np.array_equal(add_noise(volumes[1:], 1.0, seed=7, indices=[1, 2]), noisy[1:])
True

nifreeze.testing.phantoms.noise_sigma(bzero: ndarray, white_matter: ndarray, snr: float) → float

Standard deviation of the noise in each channel for a given b=0 SNR.

Parameters:

• bzero (ndarray) – A noise-free b=0 volume.

• white_matter (ndarray) – Boolean mask of white matter, where the SNR is defined.

• snr (float) – Mean white-matter b=0 signal over the noise standard deviation.

Returns:

The standard deviation of the real and of the imaginary noise.

Return type:

float

Examples

>>> noise_sigma(np.full((2, 2, 2), 30.0), np.ones((2, 2, 2), dtype=bool), 10.0)
3.0