Helper Module for Deep Learning.

Module that provides functions to prepare the Brats dataset.

class pynet.datasets.connectome.ConnectomeInjury(base_filename, n_injuries=2, signature_seed=333)[source]¶

__init__(base_filename, n_injuries=2, signature_seed=333)[source]¶

Use to create synthetic injury data.

generate_injury(n_samples=100, noise_weight=0.125)[source]¶

Return n_samples of synthetic injury data and corresponding injury strength.

static generate_injury_signatures(X_mn, n_injuries, r_state)[source]¶

Generates the signatures that represent the underlying signal in our synthetic experiments. d : (integer) the size of the input matrix (assumes is size dxd)

static load_base_connectome(file_name, verbose=False)[source]¶

Loads the connectome that serves as the base of the synthetic data.

static sample_injury_strengths(n_samples, X_mn, A, B, noise_weight)[source]¶

Returns n_samples connectomes with simulated injury from two sources.

pynet.datasets.connectome.apply_injury_and_noise(X, Sig_A, weight_A, Sig_B, weight_B, noise_weight)[source]¶

Returns a symmetric, signed, noisy, adjacency matrix with simulated injury from two sources.

pynet.datasets.connectome.fetch_connectome(datasetdir, n_samples=112, seed=333)[source]¶

Fetch/prepare the Connectome injury dataset for pynet.

Refactoring of ann4brains.synthetic.injury.ConnectomeInjury.

To simulate realistic synthetic examples, a mean connectome of preterm infant data is perturbed by a simulated focal brain injury using a local signature pattern. Two focal injury signatures are applied on two injury regions. These two regions are chosen as the two rows with the highest median responses in order to simulate injury to important regions (i.e., hubs) of the brain.

Parameters

datasetdir: str

the dataset destination folder.

n_samples: int, default 112

the number of samples.

seed: int, default None

use an int to make the randomness deterministic.

Returns

injury: ConnectomeInjury

object used to create synthetic injury data.

x_train, y_train, x_test, y_test, x_valid, y_valid: array

the train/validation/test datasets.

pynet.datasets.connectome.get_k_strongest_regions(X, k, verbose=False)[source]¶

Return the k regions (matrix columns) with the highest median values.

pynet.datasets.connectome.get_symmetric_noise(m, n)[source]¶

Return a random noise image of size m x n with values between 0 and 1.

pynet.datasets.connectome.simulate_injury(X, weight_A, sig_A, weight_B, sig_B)[source]¶