FIB data cleaner tool

Why?

Focused Ion Beam milling based tomography is a powerful technique producing huge datasets. Widely available tools (not this) work quite satisfactory when dataset is homogenous and there is limited drift between the taken slice images. But what if FIB-SEM geometric limitations and requirements of results cause pathological cases where automatic alignement is not working (i.e. surface not perfectly perpendicular to FIB; subdivided takes on acquision with differing brightness/contrast; A need to sense a very subtile composition difference: High e-beam Current causing drifts)?

Capabilities

This tool is intended to be an intermediate tool in FIB data handling workflow. It uses Hyperspy to load and save data (input/output) and load up and save metadata. The stack of images should be constructed before with Hyperspy (or ImageJ, or some SEM OEM software gives prestacked TIFF, which can be loaded directly). Currently it is possible to use built-in QtConsole to use hyperspy to load and constructruct stack of images from within the software. This tool provides posibilities to:

• do simple lossless pixel-shift (nupy.roll) based alignment of slices
• correct pathological "skewed"-charging effect with affine transformation
• affine transformation can be defined by 3x3 matrix, or/ and using a 3 draggable pivot points
• can lock a single slice, make it semi-transparent and compare with other slices
• can shift single or more selectable amount of slices simultaniously.
• PolyLine ROI tool for fast course alignment in one direction (vertical or horizontal)
• shows the views in (x,y) (y,z) and (x,z) perpendicular plainar views of the dataset. All views are updated after any slice shift or transformation.
• shift instrucitons and affine transformation matric'es can be saved and loaded and applied to the initial dataset from simple json file. This way the raw data can be left unmodified. Albeit saving the modified (shifted/cropped/transformed) dataset as a new file will reset shift and transformation instructions and any new shift and/or transformation definitions should be applied no more to the raw/initial dataset but to such new modified copy of the data.
• has built-in simple normalisation (normalizes intensities to the ROI selected at x , y, to all z)
• has included QtConsole, which allows to use other Hyperspy functions.
• Has GUI for consolidating stacked original metadata into arrays for persistant saving under original metadata of Hyperspy Signal.
• Has single button to strip away stacked original metadata, which otherwise would be converted during saving into other types, plus if not removed, it would extend loading and saving time few order of magnitude.
• posibility to blacklist slices (i.e. when FIB is not milling away the cube in sequence of images (common situation at starting up milling, or restarting/continuing after drift correction)) and remove when cropping.
• export data cube as separate images for every slice, either as "raw"-like or with LUT applied.

Requirements

Currently this is not a stand-alone application, and requires functioning python (preferably virtual) environment (conda/mamba, venv or anything similar).

• GUI is based on PyQt5 and for fast Graphical interaction uses pyqtgraph library.
• Requires hyperspy to load and save datasets, stack slices and its metadata.
• requires opencv for affine transformations and calculation of transformation matrix from sets of 3 original and moved points.

When having a working hyperspy environmet it is easy to meet above requiriments by conda install or pip install the pyqtgraph library, and pip install opencv-contrib-python-headless. Installing using pip should take care of listed dependencies.

Installation

Currently it is possible to install it with pip install . in directory where it is downloaded and extracted (it uses pyproject.toml and setuptools, there is no need for setup.py).

Launching

After the installation The software is available under FIBDataCleaner package and can be launc as stand alone or imported and used from within script or jupyter notebook.

As stand alone

It can be launched from working CLI of virtual environment simply typing fibdatacleaner. It loads current python kernel to embedded QtConsole, where hyperspy, numpy the GUI (as app) and pyqtgraph are preloaded. It allows to access hyperspy functionality from within software (i.e. more advanced loading of files). Embedded Console is inside QDockingWidget which can be deatched and moved i.e. to other monitor.

Load as library and launch from within Jupyter notebook or QtConsole.

This way it can be launched from hyperspy notebook, extending allready developed workflow. To launch it at first %gui qt5 needs to be executed in jupyter notebook if was not already. Then:

from FIBDataCleaner.fdc import FIBSliceCorrector

app = FIBSliceCorrector()
app.show()

It is possible to load 3D data cube then with app.load_hspy_signal. When launched from Jupyter notebook (or from external QtConsole), app disables embedded QtConsole.

etc

This tool does not use isig or inav from HyperSpy signals, as pyqtgraph ImageView expects numpy array. Direct access of numpy arrays allows to visualise the slices much faster than what HyperSpy's matplotlib-based visualisation is able to achieve. There was consideration at some part to go for RossetaSciIO as it has much lesser requiriments than HyperSpy. However Hyperspy provides some nice features like signal stacking. And while its metadata handling is questionable, the functionality is already there, and work-around its clumsinness was not so hard to implement.

Future considerations:

• opencv has some interesting and useful filters (i.e. Bilateral), which could come handy when dealing with noisy FIB-SEM data.
• pyqtgraph has some interesting 3D capabilities and it is worth to explore (that would bring in pyopengl as requiriments.
• alternativelly the vispy lib is worth of exploring for 3D visualisation.
• package this as library in pypi