Automatic and manual segmentation pipeline¶

[1]:

import brainlit
from brainlit.utils.session import NeuroglancerSession
from brainlit.utils.Neuron_trace import NeuronTrace
from brainlit.algorithms.generate_fragments import adaptive_thresh
import napari
from napari.utils import nbscreenshot
%gui qt5

/opt/buildhome/python3.7/lib/python3.7/site-packages/nilearn/datasets/__init__.py:96: FutureWarning: Fetchers from the nilearn.datasets module will be updated in version 0.9 to return python strings instead of bytes and Pandas dataframes instead of Numpy arrays.
  "Numpy arrays.", FutureWarning)

Find valid segments¶

In this cell, we set up a NeuroglancerSession object. Since segmentation ID numbers are not in order, we print out a list of valid IDs in some range id_range. Most segment IDs are in range(300), additionally, segments 999 and 1000 are available.

[2]:

"""
# Optional: Print the IDs of segments in Neuroglancer
url = "s3://open-neurodata/brainlit/brain1"
ngl_skel = NeuroglancerSession(url+"_segments", mip=1,use_https=False)
working_ids = []
id_range = 14
for seg_id in range(id_range):
    try:
        segment = ngl_skel.cv.skeleton.get(seg_id)
        working_ids.append(seg_id)
    except:
          pass
print(working_ids)
"""

[2]:

'\n# Optional: Print the IDs of segments in Neuroglancer\nurl = "s3://open-neurodata/brainlit/brain1"\nngl_skel = NeuroglancerSession(url+"_segments", mip=1,use_https=False)\nworking_ids = []\nid_range = 14\nfor seg_id in range(id_range): \n    try:\n        segment = ngl_skel.cv.skeleton.get(seg_id)\n        working_ids.append(seg_id)\n    except:\n          pass\nprint(working_ids)\n'

Download SWC information¶

Download the information contained in a SWC file for labelled vertices of a given seg_id at a valid mip from AWS.

[3]:

"""
seg_id = 13
mip = 2
s3_trace = NeuronTrace(url+"_segments", seg_id, mip)
df = s3_trace.get_df()
df['sample'].size # the number of vertex IDs [1, 2, ..., df['sample'].size]
"""

[3]:

'\nseg_id = 13\nmip = 2\ns3_trace = NeuronTrace(url+"_segments", seg_id, mip)\ndf = s3_trace.get_df()\ndf[\'sample\'].size # the number of vertex IDs [1, 2, ..., df[\'sample\'].size]\n'

[4]:

"""
print(df)
"""

[4]:

'\nprint(df)\n'

Select vertices¶

Select a subset of the vertices in the downloaded SWC to view and segment.

[5]:

"""
subneuron_df = df[0:5] # choose vertices to use for the subneuron
vertex_list = subneuron_df['sample'].array
print(vertex_list)
"""

[5]:

"\nsubneuron_df = df[0:5] # choose vertices to use for the subneuron\nvertex_list = subneuron_df['sample'].array \nprint(vertex_list)\n"

Download the Volume¶

Download the volume containing the specified vertices.

[6]:

"""
ngl = NeuroglancerSession(url, mip=mip)
buffer = 10
img, bounds, vox_in_img_list = ngl.pull_vertex_list(seg_id, vertex_list, buffer = buffer, expand = True)
"""

[6]:

'\nngl = NeuroglancerSession(url, mip=mip)\nbuffer = 10\nimg, bounds, vox_in_img_list = ngl.pull_vertex_list(seg_id, vertex_list, buffer = buffer, expand = True)\n'

Plot¶

[7]:

"""
# Reference: https://github.com/NeuroDataDesign/mouselit/blob/master/bijan/mouse_test/final%20notebook.ipynb
def napari_viewer(img, labels=None, shapes=None, label_name="Segmentation"):
    viewer = napari.view_image(np.squeeze(np.array(img)))
    if labels is not None:
        viewer.add_labels(labels, name=label_name)
    if shapes is not None:
        viewer.add_shapes(data=shapes, shape_type='path', edge_color='blue', name='Skeleton')
    return viewer
"""

[7]:

'\n# Reference: https://github.com/NeuroDataDesign/mouselit/blob/master/bijan/mouse_test/final%20notebook.ipynb\ndef napari_viewer(img, labels=None, shapes=None, label_name="Segmentation"):\n    viewer = napari.view_image(np.squeeze(np.array(img)))\n    if labels is not None:\n        viewer.add_labels(labels, name=label_name)\n    if shapes is not None:\n        viewer.add_shapes(data=shapes, shape_type=\'path\', edge_color=\'blue\', name=\'Skeleton\')\n    return viewer\n'

Let’s take a look at the downloaded volume. Napari will open in a new window.

[8]:

"""
viewer = napari.Viewer(ndisplay=3)
viewer.add_image(img)
nbscreenshot(viewer)
"""

[8]:

'\nviewer = napari.Viewer(ndisplay=3)\nviewer.add_image(img)\nnbscreenshot(viewer)\n'

[9]:

"""
n=napari_viewer(img)
"""

[9]:

'\nn=napari_viewer(img)\n'

[10]:

"""
import inspect
a = repr(n)
print(a)

b = repr(n).find(('napari.viewer.Viewer'))
print(b)
"""

[10]:

"\nimport inspect\na = repr(n)\nprint(a)\n\nb = repr(n).find(('napari.viewer.Viewer'))\nprint(b)\n"

[11]:

"""
n.window.close()
"""

[11]:

'\nn.window.close()\n'

[12]:

#We get a `corrected_subneuron_df` that contains `(x,y,z)` coordinates within the downloaded volume for the vertices in the SWC.

[13]:

"""
import inspect
a = repr(n)
print(a)

b = repr(n).find(('napari.viewer.Viewer'))
print(b)
"""

[13]:

"\nimport inspect\na = repr(n)\nprint(a)\n\nb = repr(n).find(('napari.viewer.Viewer'))\nprint(b)\n"

[14]:

#We get a `corrected_subneuron_df` that contains `(x,y,z)` coordinates within the downloaded volume for the vertices in the SWC.

[15]:

"""
import inspect
a = repr(n)
print(a)

b = repr(n).find(('napari.viewer.Viewer'))
print(b)
"""

[15]:

"\nimport inspect\na = repr(n)\nprint(a)\n\nb = repr(n).find(('napari.viewer.Viewer'))\nprint(b)\n"

[16]:

# We get a `corrected_subneuron_df` that contains `(x,y,z)` coordinates within the downloaded volume for the vertices in the SWC.

[17]:

"""
import inspect
a = repr(n)
print(a)

b = repr(n).find(('napari.viewer.Viewer'))
print(b)
"""

[17]:

"\nimport inspect\na = repr(n)\nprint(a)\n\nb = repr(n).find(('napari.viewer.Viewer'))\nprint(b)\n"

[18]:

# We get a `corrected_subneuron_df` that contains `(x,y,z)` coordinates within the downloaded volume for the vertices in the SWC.

[19]:

"""
transpose = vox_in_img_list.T
vox_in_img_list_t = transpose.tolist()

corrected_subneuron_df = s3_trace.generate_df_subset(list(vox_in_img_list_t), subneuron_start = 0, subneuron_end = 5)
print(corrected_subneuron_df)
"""

[19]:

'\ntranspose = vox_in_img_list.T\nvox_in_img_list_t = transpose.tolist()\n\ncorrected_subneuron_df = s3_trace.generate_df_subset(list(vox_in_img_list_t), subneuron_start = 0, subneuron_end = 5)\nprint(corrected_subneuron_df)\n'

Convert the SWC to a graph and print some information about the graph.

[20]:

"""
G = s3_trace._df_to_graph(df_voxel=corrected_subneuron_df)
print('Number of nodes:', len(G.nodes))
print('Number of edges:', len(G.edges))
print('Sample 1 coordinates (x,y,z):', G.nodes[1])
paths = s3_trace._graph_to_paths(G)
print('Number of paths:', len(paths))
"""

[20]:

"\nG = s3_trace._df_to_graph(df_voxel=corrected_subneuron_df)\nprint('Number of nodes:', len(G.nodes))\nprint('Number of edges:', len(G.edges))\nprint('Sample 1 coordinates (x,y,z):', G.nodes[1])\npaths = s3_trace._graph_to_paths(G)\nprint('Number of paths:', len(paths))\n"

We can display the SWC on the Volume

[21]:

"""
%gui qt
napari_viewer(img, shapes=paths)
nbscreenshot(viewer)
"""

[21]:

'\n%gui qt\nnapari_viewer(img, shapes=paths)\nnbscreenshot(viewer)\n'

Automatically segment the neuron¶

We start by converting the seed points to a format used by the thresholding.

[22]:

"""
seed = [adaptive_thresh.get_seed(sample)[1] for sample in vox_in_img_list]
print(seed)
"""

[22]:

'\nseed = [adaptive_thresh.get_seed(sample)[1] for sample in vox_in_img_list]\nprint(seed)\n'

Next, we compute a confidence-connected threshold segmentation.

[23]:

"""
labels = adaptive_thresh.confidence_connected_threshold(img, seed, num_iter=1, multiplier=0.5)
"""

[23]:

'\nlabels = adaptive_thresh.confidence_connected_threshold(img, seed, num_iter=1, multiplier=0.5)\n'

We can display the volume, SWC, and segmentation in Napari.

[24]:

"""
%gui qt
viewer = napari_viewer(img, labels=labels, shapes=paths, label_name="Confidence-Connected Threshold")
nbscreenshot(viewer)
"""

[24]:

'\n%gui qt\nviewer = napari_viewer(img, labels=labels, shapes=paths, label_name="Confidence-Connected Threshold")\nnbscreenshot(viewer)\n'

Steps to Manually Edit Labels¶

Labels can be manually edited following these steps:

Ensure Napari is in 2D-slice viewing, not 3D view. (The second button from the bottom left)
Click the image layer and adjust the contrast limits as desired.
Click the “Confidence-Connected Threshold Layer”
Click the paintbrush tool and adjust the brush size. Ensure that “label” is set to 1 to paint and 0 to erase.
Click and drag on the image to adjust labels. Changes are saved automatically, and CMD-Z to undo is supported.

Extract the manual labels for uploading.

[25]:

#manual_labels = viewer.layers['Confidence-Connected Threshold'].data

Upload the segmentation to AWS.

[26]:

#%%capture
#ngl_upload = NeuroglancerSession(url+"_seg", mip=mip)
#ngl_upload.push(manual_labels, bounds);

Confirm that the upload was successful. It was!

[27]:

#downloaded_labels = ngl_upload.pull_bounds_seg(bounds)

[28]:

#print(np.all(manual_labels == downloaded_labels))