

Plugins

Plugins can perform post-processing on reconstructed slices.

Developing a plugin

Developing a post-processing plugin is as easy as implementing a single Python function that takes a 2D numpy array (the reconstructed slice), and returns a 2D numpy array (the postprocessed slice). An example plugin looks like this.

import numpy as np
import slicerecon
from skimage import filters

def callback(shape, xs, _):
    # The Otsu implementation does not accept only zeros, so we pass through
    if not np.array(xs).any():
        return [shape, xs]

    # Reshape into an image
    xs = np.array(xs).reshape(shape)

    # Compute a threshold using Otsu's method
    val = filters.threshold_otsu(xs)

    # Threshold the image accordingly
    xs[xs <= val] = 0.0
    xs[xs > val] = 1.0

    return [shape, xs.ravel().tolist()]

# Host a plugin on the default port, with the default RECAST3D endpoint
p = slicerecon.plugin("tcp://*:5652", "tcp://localhost:5555")

# Register the callback
p.set_slice_callback(callback)

# Start the plugin
p.listen()

This plugin listens to incoming SliceData packets on port 5652, and connects to a visualization software (or another plugin) listening on port 5555. These are the default values. If you use the standard slicerecon_server program, connecting the Python plugin is as easy as passing --pyplugin as a flag.

This plugin computes a simple segmentation based on a threshold computed by Otsu’s method.

Testing your plugin

1. Start RECAST3D:

recast3d

2. Start slicerecon_server, e.g.:

slicerecon_server --slice-size 512 --pyplugin

3. Run your plugin, e.g.:

python plugin.py

4. Stream projection data to the slicerecon_server, e.g.:

python slicerecon_push_flexdata.py [path_to_data] --sample 2

If you make a change to plugin.py, you can restart it without touching the other components (recast3d, slicerecon) and without restreaming the projection data. Simply request a new reconstruction by modifying a slice (e.g. by clicking on one of the slices in RECAST3D) to see the result.

Example: Otsu thresholding

The example above should give the following visual result.

Dataset: A cone beam scan of a rat skull: doi:10.5281/zenodo.1164088