ETSpy Demo¶

This notebook demonstrates the basic use and functionality of the ETSpy package.

It covers:

• Loading simulated data

• Basic plotting of tilt series data

• Reconstruction of single slices of simulated data

• Reconstruction of the full simluated dataset

• Saving reconstructed data

If you are viewing this demo on the ETSpy documentation website, it will not be interactive. We recommend downloading the latest version of the demo from Github and running it on your local machine.

Imports¶

For interactive plotting, we set the Matplotlib backend to widget.

In addition to the etspy package itself we also need to import:

• pyplot from matplotlib

• hyperspy

Finally, we also need to import the datasets module of etspy in order to load the simluated data we will be working with.

# %matplotlib inline

## Set to "ipympl" for interactive plots
%matplotlib ipympl

import matplotlib.pyplot as plt
import hyperspy.api as hs

import etspy.api as tomo
from etspy import datasets as ds

Simulated Catalyst Tilt Series¶

Load Data¶

• Read simulated catalyst tilt series as a TomoStack object

• Each image is shifted randomly to simulate specimen motion during tilt series acquisition

stack = ds.get_catalyst_data(misalign=True)

Metadata about the tilt series is contained in the Tomography node:

stack.metadata.Tomography

Information about the shifts and the tilts for each projection in the series are contained in the shifts and tilts attributes as HyperSpy signal objects, respectively:

stack.tilts

<TomoTilts, title: Image tilt values, dimensions: (90|1)>

stack.tilts.axes_manager

stack.shifts

<TomoShifts, title: Image shift values, dimensions: (90|2)>

# show the shifts of the first five projections (in Y/X order):
stack.shifts.data[:5, :]

array([[ 4.46601307, -4.83013074],
       [ 2.50940259,  4.3421752 ],
       [-3.72235869,  2.2149008 ],
       [ 4.26289602,  2.07969516],
       [ 0.77610969,  3.44610051]])

Rebin the data¶

• To speed things up, we can rebin the dataset by a factor of 2 in the X and Y dimensions

• Uses the underlying Hyperspy rebin method which all TomoStack’s inherit from their parent Signal2D class

rebin = stack.rebin(scale=[1, 2, 2])
rebin

<TomoStack, title: , dimensions: (90|300, 300)>

Browse the Rebinned Data¶

• Uses Hyperspy’s plotting functionality

• Plot and view the full series interactively

• Plot a max image of the dataset which emphasizes the misalignment of the stack

rebin.plot(cmap='inferno')

rebin.max().plot(cmap='inferno')

Stack Registration¶

• Five methods available:

  • Phase correlation (OpenCV)

  • Enhanced correlation coefficient (OpenCV)

  • StackReg (pystackreg)

  • Center of Mass (T. Sanders et al. doi:10.13140/RG.2.2.33492.60801)

  • Combined Center of Mass and Common-Line Method (UCLA, Scott et al. doi:10.1038/nature10934)

reg = rebin.stack_register('StackReg')

# compare the non-registered and registered max-pixel images:
hs.plot.plot_images(
    [rebin.max(), reg.max()],
    per_row=1,
    label=["Before registration", "After registration"],
    cmap='inferno'
)
plt.gcf().set_size_inches((4,8))
plt.tight_layout()

Reconstruction¶

Plot single image alongside the sinogram of the central slice¶

sino = reg.extract_sinogram(150)

ax = hs.plot.plot_images([sino, reg.inav[45]], cmap='inferno', per_row=1, label=['Sinogram','FBP Reconstruction'])
ax[1].axhline(300, linestyle='--', color='red')
plt.gcf().set_size_inches((6,6))
plt.tight_layout()

Filtered Backprojection of Single Slice (CPU)¶

recFBP = reg.isig[150:151, :].reconstruct('FBP', cuda=False)

[########################################] | 100% Completed | 205.71 ms

ax = hs.plot.plot_images([sino, recFBP], cmap='inferno', per_row=1, label=['Sinogram','FBP Reconstruction'])
plt.gcf().set_size_inches((6,6))
plt.tight_layout()

SIRT Error Analysis¶

• Perform a SIRT reconstruction and calculate the L2-norm between the forward-projection of each result and the input sinogram

• The reconstruction result is also saved at each iteration for viewing

• Note: Currently, the error output differs between CUDA- and CPU-based reconstructions. Cause is unknown.

SIRTStack, SIRTerror = reg.isig[150:151, :].recon_error(iterations = 500)
SIRTStack.axes_manager[2].name = 'z'

100%|██████████████████████████████████████████████████████████████████| 500/500 [00:00<00:00, 688.31it/s]

SIRTStack.plot(navigator=SIRTerror, cmap='inferno')

ax = hs.plot.plot_images([SIRTStack.inav[0],SIRTStack.inav[19],SIRTStack.inav[49],
                         SIRTStack.inav[99], SIRTStack.inav[249], SIRTStack.inav[499]],
                         cmap='inferno', per_row=2,
                         axes_decor='off',
                         label=['Iterations: 1','Iterations: 20','Iterations: 50',
                                'Iterations: 100','Iterations: 250','Iterations: 500'])
plt.gcf().set_size_inches(6,9)
plt.tight_layout()

Reconstruct the entire binned stack¶

• Takes less than 30 seconds if using GPU-acceleration (will fail if CUDA tooling not installed)

recFBP = reg.reconstruct('FBP', cuda=True)
recSIRT = reg.reconstruct('SIRT', iterations=100, constrain=True, cuda=True)

100%|██████████████████████████████████████████████████████████████████| 300/300 [00:00<00:00, 415.34it/s]
100%|███████████████████████████████████████████████████████████████████| 300/300 [00:18<00:00, 16.42it/s]

recFBP.plot(vmax=2000, cmap='inferno')

recSIRT.plot(vmax=2000, cmap='inferno')

recSIRT.swap_axes(0,2).plot(vmax=2000, cmap='inferno')

Save the Reconstruction Results¶

• Hyperspy-compatible HDF5 is the output format

recFBP.save('FBP_Reconstruction.hdf5', overwrite=True, file_format='HSPY')
recSIRT.save('SIRT_Reconstruction_100.hdf5', overwrite=True, file_format='HSPY')

Experimental Needle-shaped Sample Tilt Series¶

• FIB-milled specimen from NIST SRM-2135c

  • Ni/Cr mutli-layer thin film on silicon substrate

stack = ds.get_needle_data()

stack.plot(navigator='slider')

Stack Registration¶

• Five methods available:

  • Phase correlation (OpenCV)

  • Enhanced correlation coefficient (OpenCV)

  • StackReg (pystackreg)

  • Center of Mass (T. Sanders et al. doi:10.13140/RG.2.2.33492.60801)

  • Combined Center of Mass and Common-Line Method (UCLA, Scott et al. doi:10.1038/nature10934)

reg = stack.stack_register('StackReg')

reg.plot(navigator='slider')

f = reg.test_align(slices=[10,100,140])

Tilt Axis Alignment¶

• Uses the center of mass method (CoM)

• Motion of specimen is fit to that expected for an ideal cylinder

• Fit is performed for three different sinograms

• The result is used to estimate the tilt axis shift and rotation

ali = reg.tilt_align('CoM')

f = ali.test_align(slices=[10,100,140])

ali.plot()

Reconstruction¶

# Takes about a minute on a GPU-accelerated system
rec = ali.reconstruct('SIRT', 500, constrain=True)

100%|███████████████████████████████████████████████████████████████████| 256/256 [01:00<00:00,  4.26it/s]

rec.inav[10:205].plot(cmap='inferno', vmin=50, vmax=1200)

rec.isig[:,90:165].swap_axes(2,0).plot(cmap='inferno', vmin=50)