tracking-3d

Feature finding in 3D confocal images¶

Confocal microscopy is a techique that is able to make image slices of your sample. New developments (spinning disk scanning, resonant mirrors) have made this technique sufficiently fast for 3D tracking. Commonly, the sample is mounted on a piezo stage that is able to quickly scan the sample in Z direction. In this way, sequences of Z pictures can be made. By repeating this, the sample can be followed real-time in 3D.

Because of the different techniques used for XY and Z resolution, pixel sizes in these directions are mostly not equal. It is easy to extend the method of Crocker & Grier to 3 dimensions with different feature sizes. Trackpy can do this from version v0.3; the only thing you need to do is load your images as 3 dimensional numpy arrays and supply a tuple value for diameter, describing feature size in each dimension.

First, we initalize IPython and load the required packages. PIMS and Trackpy v0.3 are required.

Setup IPython, pims, trackpy¶

In [1]:

%matplotlib inline
import matplotlib as mpl
import matplotlib.pyplot as plt
mpl.rc('figure',  figsize=(10, 6))
import numpy as np
import pandas as pd
from pandas import DataFrame, Series  # for convenience

import pims
import trackpy as tp

import warnings
warnings.filterwarnings("ignore", module="matplotlib")

Import 3D images¶

We make use of ImageSequenceND. This reader class searches for files matching a certain search string. The filename should contain indentifiers for each axis (dimension). For example "demo_t001_z001.png" contains identifiers t and z. We set the returned image axes to zyx and the iterating axis to t. In this way we get 3D images indexed by t.

For this tutorial, the images are inside a zipfile. You can extract the images from the specified archive to check them out.

In [2]:

frames = pims.ImageSequenceND(r'../sample_data/pmma_colloids.zip', axes_identifiers = ['z', 't'])
frames.bundle_axes = ['z', 'y', 'x']
frames.iter_axes = 't'
frames

Out[2]:

<ImageSequenceND>
Source: /Users/charles/codes/trackpy-examples/sample_data/pmma_colloids.zip
Axes: 4
Axis 'z' size: 25
Axis 't' size: 33
Axis 'y' size: 128
Axis 'x' size: 128
Pixel Datatype: uint8

The time-index of the first frame is always 0, regardless of the index in the filename. We can use pims 3d scrollable stack feature to display a 3D image. Use the scroll wheel to scroll through the focal planes.

This feature will not work online, download the example and run it yourself to checkout the scrollable stack feature.

In [3]:

frames[0]

Out[3]:

Feature finding¶

Now we can use trackpy.locate to generate a table (a pandas DataFrame) of the feature coordinates. locate only requires a feature diameter. This needs to be an odd integer and can be specified for each dimension seperately. The order of dimensions is the same as the order of the image array, typically (z, y, x).

In [4]:

features = tp.locate(frames[0], diameter=(5, 9, 9))
features.head()  # displays first 5 rows

Out[4]:

	x	y	z	mass	size_x	size_y	size_z	ecc	signal	raw_mass	ep_x	ep_y	ep_z	frame
0	92.767466	9.071401	2.012498	7546.110185	2.860761	2.665710	1.405736	NaN	119.948425	11471.0	0.011464	0.011464	0.005139	0
1	54.699074	12.227642	1.913505	4283.319561	2.822252	2.610693	1.350248	NaN	77.386081	7476.0	0.017994	0.017994	0.008067	0
2	67.940818	36.920317	2.033795	6039.499928	2.930073	2.924657	1.355519	NaN	83.190037	9711.0	0.013646	0.013646	0.006117	0
3	4.873056	37.946693	1.959903	5162.618902	2.897005	2.790764	1.331927	NaN	81.739048	8632.0	0.015448	0.015448	0.006925	0
4	58.815615	80.025063	2.059816	5191.155019	2.814045	2.593217	1.386939	NaN	98.183590	7986.0	0.016774	0.016774	0.007520	0

Because we specified different diameters for each dimension, trackpy we return size and ep (static error) for each dimension seperately. We can plot the feature locations on top of the original picture. You can see that there are no features found at the edges. This is because these particles partly fall out of the measurement box.

In [5]:

tp.annotate3d(features, frames[0])

Out[5]: