FLIM Workflow Example

FLIM Workflow Example#

This notebook show an example of how to automate the phasor plot generation and clustering in order to achieve a segmentation result based on lifetime differences.

import napari
import hdbscan
import pandas as pd
import napari_flim_phasor_plotter as flim_plot

from napari.utils import nbscreenshot
from skimage.measure import label, regionprops_table

Below are the parameters used in this workflow.

# phasor plot parameters
manual_threshold = 15
harmonic = 1
apply_median_filter = True
median_filter_iterations = 1

# hdbscan parameters
hdbscan_min_cluster_size = 40
hdbscan_min_samples = 40

# post-processing parameters
chosen_cluster_label = 20
fill_area_threshold = 100
smooth_radius = 3

Open sample data in napari#

Open napari viewer by running the cell below.

viewer = napari.Viewer()

Assistant skips harvesting pyclesperanto as it's not installed.
Assistant skips harvesting pyclesperanto_prototype as it's not installed.

And open a FLIM sample image (alternatively go to File -> Open Sample -> napari-flim-phasor-plotter -> Seminal Receptacle)

layer_raw_FLIM_image, layer_image = viewer.open_sample(plugin='napari-flim-phasor-plotter', sample='receptacle')
nbscreenshot(viewer)

Make the phasor plot#

Below we make a phasor plot from the raw FLIM image and display it in napari (alternatively go to Plugins -> napari-flim-phasor-plotter -> Calculate Phasors).

# Get widget instance
plot_maker_widget = flim_plot.widgets.make_flim_phasor_plot()
# Run plot maker widget
phasor_plot_widget, labels_layer = plot_maker_widget(
        image_layer=layer_raw_FLIM_image,
        threshold = manual_threshold,
        apply_median = apply_median_filter,
        median_n = median_filter_iterations,
        napari_viewer = viewer)
nbscreenshot(viewer)

Now, we get the features table from the recently created labels layer.

features_table = labels_layer.features
features_table

	label	G	S	pixel_x_coordinates	pixel_y_coordinates	pixel_z_coordinates	frame
0	1	0.368575	0.378603	0	0	0	0
1	2	0.368575	0.379517	1	0	0	0
2	3	0.333649	0.379517	2	0	0	0
3	4	0.360470	0.362263	3	0	0	0
4	5	0.360470	0.362263	4	0	0	0
...	...	...	...	...	...	...	...
123724	123725	0.399062	0.289760	183	511	0	0
123725	123726	0.399062	0.286353	184	511	0	0
123726	123727	0.167012	0.289760	185	511	0	0
123727	123728	0.251258	0.329581	186	511	0	0
123728	123729	0.361668	0.355341	189	511	0	0

123729 rows × 7 columns

Cluster ‘G’ and ‘S’ (with HDBSCAN)#

Here we use HDBSCAN to automatically find clusters and update the features table with the prediction results. Manual cluster selection could also be used in napari.

Alternatively, run Clustering on ‘G’ and ‘S’ with napari-clusters-plotter (Tools -> Measurement post-processing -> Clustering). Check napari-clusters-plotter repository for more details.

clustering_hdbscan = hdbscan.HDBSCAN(
    min_cluster_size=hdbscan_min_cluster_size,
    min_samples=hdbscan_min_samples,
)
# Run clustering on G and S coordinates
hdbscan_predictions = clustering_hdbscan.fit_predict(features_table[['G', 'S']])
clustering_id = "HDBSCAN_CLUSTER_ID"

We add HDBSCAN predictions to table.

features_table[clustering_id] = hdbscan_predictions
features_table

	label	G	S	pixel_x_coordinates	pixel_y_coordinates	pixel_z_coordinates	frame	HDBSCAN_CLUSTER_ID
0	1	0.368575	0.378603	0	0	0	0	-1
1	2	0.368575	0.379517	1	0	0	0	-1
2	3	0.333649	0.379517	2	0	0	0	-1
3	4	0.360470	0.362263	3	0	0	0	-1
4	5	0.360470	0.362263	4	0	0	0	-1
...	...	...	...	...	...	...	...	...
123724	123725	0.399062	0.289760	183	511	0	0	-1
123725	123726	0.399062	0.286353	184	511	0	0	-1
123726	123727	0.167012	0.289760	185	511	0	0	-1
123727	123728	0.251258	0.329581	186	511	0	0	-1
123728	123729	0.361668	0.355341	189	511	0	0	-1

123729 rows × 8 columns

And we update the layer features with the table as well as the plotter clustering options.

# update layer features table
labels_layer.features = features_table
# update clustering id list
phasor_plot_widget.update_axes_and_clustering_id_lists()
# set clustering id in dropdown
phasor_plot_widget.plot_cluster_id.setCurrentText(clustering_id)

Finally, we display clusters predictions on the phasor plot and the corresponding pixels as a new Labels layer.

phasor_plot_widget.run(
    labels_layer.features,
    phasor_plot_widget.plot_x_axis.currentText(),
    phasor_plot_widget.plot_y_axis.currentText(),
    phasor_plot_widget.plot_cluster_id.currentText(),)
# Get clusters labels layer
clusters_labels_layer = phasor_plot_widget.visualized_layer
nbscreenshot(viewer)

Post-Processing#

Extract chosen label#

From the clusters, we could choose one (or more) cluster of interest to further process segmentation results.

We use the manual_label_extract function for that.

Alternatively, use Plugins -> napari-flim-phasor-plotter -> Manual Label Extraction

extraced_labels_layer = flim_plot.widgets.manual_label_extract(clusters_labels_layer, chosen_cluster_label)
viewer.layers[-1].visible = False
viewer.add_layer(extraced_labels_layer)
nbscreenshot(viewer)

Optionally we can add the n largest cluster labels to napari as individual layers with split_n_largest_cluster_labels (also available via Plugins -> napari-flim-phasor-plotter -> Split N Largest Clusters Labels).

list_of_labels_layers = flim_plot.widgets.split_n_largest_cluster_labels(
    labels_layer=labels_layer,
    clusters_labels_layer=clusters_labels_layer,
    clustering_id=clustering_id,
    n=5,
)
# Uncomment code below to add layers to viewer
# for layer in list_of_labels_layers:
#     viewer.add_layer(layer)

Smooth chosen cluster mask#

Different morphological post-processing could be applied. We provide a general smoothing function in the plugin called smooth_cluster_mask.

Alternatively, use Plugins -> napari-flim-phasor-plotter -> Smooth Cluster Mask.

smoothed_cluster_label = flim_plot.widgets.smooth_cluster_mask(
    cluster_mask_layer=extraced_labels_layer,
    fill_area_px=fill_area_threshold,
    smooth_radius=smooth_radius,
)
smoothed_cluster_label.name = 'smoothed_cluster_label'
viewer.layers[-1].visible = False
viewer.add_layer(smoothed_cluster_label)
nbscreenshot(viewer)

Instance Segmentation and Feature Extraction#

Now, we perform instance segmentation and feature extraction using scikit-image. You can optionally:

apply connected components labeling with Tools -> Segmentation / Labeling -> Connected component labeling;
and perform features extraction with Tools -> Measurement Tables -> Object Features/Properties.

objects_labels = label(smoothed_cluster_label.data)
features_table = pd.DataFrame(regionprops_table(label_image = objects_labels, properties = ['label', 'area']))
viewer.layers[-1].visible = False
viewer.add_labels(objects_labels, features = features_table, name = 'receptacle_objects')
nbscreenshot(viewer)

features_table

	label	area
0	1	41.0
1	2	23236.0
2	3	320.0
3	4	431.0
4	5	3963.0

Optionally, we can display the table in napari with Tools > Measurement > Show Table; Select the ‘receptacle objects’ layer and click Run.

nbscreenshot(viewer)