Slide: 1

V.A. Solé - European Synchrotron Radiation Facility2009 Python for Scientific Computing Conference

The PyMca Application and Toolkit

Slide: 2

The ESRF: Just an X-Ray Source

Slide: 3

(Partial) Synchrotron Chart

Slide: 4

X-Ray Fluorescence Basics

Energy of emitted X rays is element dependent

Slide: 5

Visualization via Multi-Channel Analyzers (MCA)

Slide: 6

PyMca?

For the end users and the specific developers

Set of programs and widgets for XRF analysis

Spectrum modeling

Quantification

ROI imaging

Fit imaging via batch processing

For the general developer

Set of python modules

Data visualization

Peak search

Function fitting

Imaging of 1D data

V.A. Solé, E. Papillon, M. Cotte, Ph. Walter, J. Susini, Spectrochimica Acta B 62 (2007) 63-68

PyMca is set of software tools on its way to become a reference in XRF

Slide: 7

- First release in 2004

- Source code SVN repository available at sourceforge since 2006

- Official releases aim to provide ready-to-use binaries to end users

- External dependencies on numpy, PyQt4 and PyQwt5

- If installed, Matplotlib is used to generate high quality output

- Provides high level functional widgets that can be embedded in your own application

Generalities

Slide: 8

XRF Spectrum Analysis

Typical procedure:

1. Calibration

2. Peak identification

3. Peak area extraction

Region of interest (ROI)

Deconvolution (FIT)

4. Quantification

Documentation at http://pymca.sourceforge.net/documentation.html

Slide: 9

1. Calibration

Detailed documentation at http://pymca.sourceforge.net/documentation.html

Slide: 10

2. IdentificationTheoretical database in ASCII format

IUPAC notation whenever possible

PyMca proposes, the user decidesPyMca proposes, the user decidesPyMca proposes, the user decidesPyMca proposes, the user decides

Slide: 11

3.1 Peak area via ROI

Slide: 12

3.2 Peak area via fitting

Slide: 13

3.2 Peak area via fitting

Slide: 14

Fit configuration Dialog (I)

Slide: 15

Fit Configuration Dialog (II)

Slide: 16

Emission Lines

Allowed transitions Implemented transitions

Any transition defined

KShellRates.datKShellRates.datKShellRates.datKShellRates.dat

LShellRates.datLShellRates.datLShellRates.datLShellRates.dat

MShellRates.datMShellRates.datMShellRates.datMShellRates.dat

Slide: 17

4. Quantification (I)

Parallel beam approximationParallel beam approximationParallel beam approximationParallel beam approximation

Slide: 18

4. Quantification (II)Nominal concentration 500 ppm

Slide: 19

XRF Analysis Integration in other Applications

Integration in mxCuBE (ESRF) Integration elsewhere

Slide: 20

Is it easy to embed?

from PyMca import McaAdvancedFit

fitWindow = McaAdvancedFit.McaAdvancedFit()

fitWindow.setData(x, y)

fitWindow.show()

For the previous examples, basically one just needs 4 lines of code:

It can be used from ipython just starting it as “ipython –q4thread”

Slide: 21

Advanced Fit Batch processing

Select the input files

Select the fit configuration

Select the output directory

Select the output options

Start

Slide: 22

Output

Images in ASCII and ESRF format

• Easy to import in other programs

Individual peak contributions in ASCII

• Use your own plotting program

Fully automated HTML report

• Browse your results!

Slide: 23

+……

One spectrum for each pixel

A sum spectrum for the whole image

XRF Imaging

Slide: 24

Sb

Pb

S

Ca

We wanted to know what pigments were

used in this section of the painting

Copyright C2RMF

Slide: 25

Based on the batch generated element distribution maps …

Slide: 26

… we were able to determine the possible presence of stibnite grains (Sb2S3) embedded in a lead containing matrix.

M. Cotte, E. Welcomme, V.A. Solé, M. Salomé, M. Menu, Ph. Walter, J. Susini, Anal. Chem. 79 (2007) 6988-6994

Sulfur and antimony correlated Lead and antimony not correlated

… and their correlations as shown by the program

Slide: 27

Generic Fitting Module

Levenberg-Marquardt algorithm with constraints on fitting parameters

It accepts user defined functions

The simplest form of user function:

Slide: 28

Automatic Peak Search Routines

Slide: 29

In this example:

Stack = 101x200x2000 numpy array

20200 spectra of 2000 channels

Pixel[i, j] = numpy.sum(Stack[i, j, :])

Pixel[i, j] = numpy.sum(Stack[i, j, ch0:ch1])

1D Stack ROI Imaging

Slide: 30

1D Stack ROI Imaging

We can generate new images by moving the cursors or defining new ROIs in the table

Slide: 31

Imaging and Principal Component Analysis

Subtract the average spectrum from each spectrum of the dataset and arrange it as NpixelsxNchannels

Calculate the covariance = Data.T * Data

Get the eigenvectors of the covariance

Add the average spectrum to each row and get the projections of that dataset on each eigenvector (Npixels scalars for each eigenvector)

Reshape the projections in the original map shape to obtain the ¨Eigenimages¨

Covariance method, other ways in http:// en.wikipedia.org/wiki/Principal_components_analysis

Slide: 32

Eigenimages and Eigenvectors

Slide: 33

Eigenimages and Eigenvectors

Slide: 34

Eigenimages and Eigenvectors

Slide: 35

Eigenimages and Eigenvectors

Slide: 36

Eigenimages and Eigenvectors

Slide: 37

Eigenimages and Eigenvectors

Slide: 38

Getting the actual information

We can select a set of pixels on any of the displayed images and display the cumulative spectrum associated to those pixels.

Here we can see the average spectrum associated to the hotter pixels of the Eigenimage 02 (in red) compared to the average spectrum of the map (in black).

Slide: 39

We could have easily missed the presence of one element if we would have just analyzed the sum spectrum via ROIs.

Slide: 40

What have we done?

We have used principal components analysis to know what sample regions were worth to take a closer look.

Not bad when you have a lot of data …

This data treatment is totally generic and applicable to other methods of analysis

Slide: 41

Current Developments – HDF5 Support

- Collaboration with D. Dale from Cornell High Energy Synchrotron Source (CHESS)

- HDF5 expected to become the file format of most European synchrotrons sources

- Additional dependency on h5py http://h5py.alfven.org

- Partial support already available in current SVN sources

Slide: 42

Current Developments – 3D and 4D

-The goal is to provide ways to visualize and interact with multiparametric data

-Already available at the ESRF through an additional PyMca set of modules

-Currently evaluating the pros and cons of using MayaVi

Slide: 43

Slide: 44

Slide: 45

Conclusion

PyMca is a program as well as an XRF analysis toolkit

- Open source and distributed under the conditions of the GPLv2+

- Can be used as a fitting and visualization tool (for up to 4-dimensional data)

- Provides high level widgets based on PyQt that can be used independently or integrated into your application

- Allows you to specify a physically meaningful model which can quantitatively determine element concentrations from energy dispersive X-ray spectra

- Is implemented at many synchrotron facilities and labs around the world

- Active development is funded by the ESRF

Slide: 46

Acknowledgements

My ESRF colleagues (mainly software group and microscopy beamlines ID21 and ID22)

Ph. Walter and M. Cotte (Centre de Recherche et de Restauration des Musées de France)

Darren Dale (Cornell High Energy Synchrotron Source)

The PyMca users, for their enthusiasm and their encouragements