Datasets

We include multiple dataset loading utilities. Note that for some functions implemented in the atlalign.data module we assume that the user has the underlying raw data locally.

Nissl volume

Contains the 25 microns reference volume used within the entire project. The corresponding function is nissl_volume that returns a (528, 320, 456, 1) numpy array.

Annotation volume

Contains per pixel segmentation/annotation of the entire reference atlas. The corresponding function is annotation_volume that returns a (528, 320, 456) numpy array. Note that one can use the corresponding segmentation_collapsing_labels function to get the tree-like hierarchy of the classes.

Manual registration

This dataset comes from the labeling tool that extracts displacement fields. The corresponding function is manual_registration. Available genes:

  • Calb1

  • Calb2

  • Cck

  • Npy

  • Pvalb

  • Sst

  • Vip

from atlalign.data import manual_registration

res = manual_registration()

assert set(res.keys()) == {'dataset_id', 'deltas_xy', 'image_id', 'img', 'inv_deltas_xy', 'p'}
assert len(res['image_id']) == 278

The returned dictionary contains the following keys:

  • dataset_id - unique id of the section dataset

  • deltas_xy - array of shape (320, 456, 2) where the last dimension represents the x (resp y) deltas of the transformation

  • image_id - unique id of the section image

  • img - moving image of shape (320, 456) that was preregistered with the Allen API

  • inv_deltas_xy - same as deltas_xy but represents the inverse transformation

  • p - coronal coordinate in microns [0, 13200]

To perform the registration instantiate atlalign.base.DisplacementField using the deltas_xy and warp the img with it.

from atlalign.base import DisplacementField
from atlalign.data import manual_registration

import numpy as np

res = manual_registration()
i = 10
delta_x = res['deltas_xy'][i, ..., 0]
delta_y = res['deltas_xy'][i, ..., 1]
img_mov = res['img'][i]

df = DisplacementField(delta_x, delta_y)
img_reg = df.warp(img_mov)

For more details on atlalign.base.DisplacementField see Building Blocks.

Dummy

Artificially generated datasets.

Rectangles

Rectangles with stripes of different intensities. Corresponding function - rectangles.

Circles

Circles with inner circles of different intensities. Corresponding function - circles.

Source

source/datasets.rst